doc. Ing. Jiří Kléma, Ph.D.

Recent research has already shown that circular RNAs (circRNAs) are functional in gene expression regulation and potentially related to diseases. Due to their stability, circRNAs can also be used as biomarkers for diagnosis. However, the function of most circRNAs remains unknown, and it is expensive and time-consuming to discover it through biological experiments. In this paper, we predict circRNA annotations from the knowledge of their interaction with miRNAs and subsequent miRNA–mRNA interactions. First, we construct an interaction network for a target circRNA and secondly spread the information from the network nodes with the known function to the root circRNA node. This idea itself is not new; our main contribution lies in proposing an efficient and exact deterministic procedure based on the principle of probability-generating functions to calculate the p-value of association test between a circRNA and an annotation term. We show that our publicly available algorithm is both more effective and efficient than the commonly used Monte-Carlo sampling approach that may suffer from difficult quantification of sampling convergence and subsequent sampling inefficiency. We experimentally demonstrate that the new approach is two orders of magnitude faster than the Monte-Carlo sampling, which makes summary annotation of large circRNA files feasible; this includes their reannotation after periodical interaction network updates, for example. We provide a summary annotation of a current circRNA database as one of our outputs. The proposed algorithm could be generalized towards other types of RNA in way that is straightforward.

Background/Aim: Prediction of response to azacitidine (AZA) treatment is an important challenge in hematooncology. In addition to protein coding genes (PCGs), AZA efficiency is influenced by various noncoding RNAs (ncRNAs), including long ncRNAs (lncRNAs), circular RNAs (circRNAs), and transposable elements (TEs). Materials and Methods: RNA sequencing was performed in patients with myelodysplastic syndromes or acute myeloid leukemia before AZA treatment to assess contribution of ncRNAs to AZA mechanisms and propose novel disease prediction biomarkers. Results: Our analyses showed that lncRNAs had the strongest predictive potential. The combined set of the best predictors included 14 lncRNAs, and only four PCGs, one circRNA, and no TEs. Epigenetic regulation and recombinational repair were suggested as crucial for AZA response, and network modeling defined three deregulated lncRNAs (CTC-482H14.5, RP11-419K12.2, and RP11-736I24.4) associated with these processes. Conclusion: The expression of various ncRNAs can influence the effect of AZA and new ncRNA-based predictive biomarkers can be defined.

Gene co-expression networks have frequently been used for functional annotation. In these networks, an unknown gene is annotated with terms that have already been associated with genes whose expression profiles t end to correlate with the expression profile of the unknown gene. Despite the biological plausibility of this principle referred to as guilt-by-association, its applicability has not been thoroughly experimentally verified yet. In our paper, we formulate several statistical hypotheses concerning the principle and test them on a representative expression dataset. We demonstrate that gene annotation carried out with co-expression networks clearly outperforms random annotation and improves with increasing sample size and the knowledge of gene co-location. Eventually, we discuss the practical significance of this way of functional annotation.

We aimed to identify the variables that modify levels of oxidatively damaged DNA and lipid peroxidation in subjects living in diverse localities of the Czech Republic (a rural area, a metropolitan locality, and an industrial region). The sampling of a total of 126 policemen was conducted twice in two sampling seasons. Personal characteristics, concentrations of particulate matter of aerodynamic diameter <2.5 mu m and benzo[a]pyrene in the ambient air, activities of antioxidant mechanisms (superoxide dismutase, catalase, glutathione peroxidase, and antioxidant capacity), levels of pro-inflammatory cytokines (TNF-alpha, IL-1 beta, and IL-6), concentrations of persistent organic pollutants in blood plasma, and urinary levels of polycyclic aromatic hydrocarbon metabolites were investigated as parameters potentially affecting the markers of DNA oxidation (8-oxo-7,8-dihydro-2 '-deoxyguanosine) and lipid peroxidation (15-F2t-isoprostane). The levels of oxidative stress markers mostly differed between the localities in the individual sampling seasons. Multivariate linear regression analysis revealed IL-6, a pro-inflammatory cytokine, as a factor with the most pronounced effects on oxidative stress parameters. The role of other variables, including environmental pollutants, was minor. In conclusion, our study showed that oxidative damage to macromolecules was affected by processes related to inflammation; however, we did not identify a specific environmental factor responsible for the pro-inflammatory response in the organism.

Patients with lower-risk myelodysplastic syndromes (LR-MDS) have a generally favorable prognosis; however, a small proportion of cases progress rapidly. This study aimed to define molecular biomarkers predictive of LR-MDS progression and to uncover cellular pathways contributing to malignant transformation. The mutational landscape was analyzed in 214 LR-MDS patients, and at least one mutation was detected in 137 patients (64%). Mutated RUNX1 was identified as the main molecular predictor of rapid progression by statistics and machine learning. To study the effect of mutated RUNX1 on pathway regulation, the expression profiles of CD34 + cells from LR-MDS patients with RUNX1 mutations were compared to those from patients without RUNX1 mutations. The data suggest that RUNX1-unmutated LR-MDS cells are protected by DNA damage response (DDR) mechanisms and cellular senescence as an antitumor cellular barrier, while RUNX1 mutations may be one of the triggers of malignant transformation. Dysregulated DDR and cellular senescence were also observed at the functional level by detecting gamma H2AX expression and beta-galactosidase activity. Notably, the expression profiles of RUNX1-mutated LR-MDS resembled those of higher-risk MDS at diagnosis. This study demonstrates that incorporating molecular data improves LR-MDS risk stratification and that mutated RUNX1 is associated with a suppressed defense against LR-MDS progression.

Background Ubiquitin ligases (Ub-ligases) are essential intracellular enzymes responsible for the regulation of proteome homeostasis, signaling pathway crosstalk, cell differentiation and stress responses. Individual Ub-ligases exhibit their unique functions based on the nature of their substrates. They create a complex regulatory network with alternative and feedback pathways to maintain cell homeostasis, being thus important players in many physiological and pathological conditions. However, the functional classification of Ub-ligases needs to be revised and extended. Methods In the current study, we used a novel semantic biclustering technique for expression profiling of Ub-ligases and ubiquitination-related genes in the murine gastrointestinal tract (GIT). We accommodated a general framework of the algorithm for finding tissue-specific gene expression clusters in GIT. In order to test identified clusters in a biological system, we used a model of epithelial regeneration. For this purpose, a dextran sulfate sodium (DSS) mouse model, following with in situ hybridization, was used to expose genes with possible compensatory features. To determine cell-type specific distribution of Ub-ligases and ubiquitination-related genes, principal component analysis (PCA) and Uniform Manifold Approximation and Projection technique (UMAP) were used to analyze the Tabula Muris scRNA-seq data of murine colon followed by comparison with our clustering results. Results Our established clustering protocol, that incorporates the semantic biclustering algorithm, demonstrated the potential to reveal interesting expression patterns. In this manner, we statistically defined gene clusters consisting of the same genes involved in distinct regulatory pathways vs distinct genes playing roles in functionally similar signaling pathways. This allowed us to uncover the potentially redundant features of GIT-specific Ub-ligases and ubiquitination-related genes. Testing the statistically obtained results on the mouse model showed that genes clustered to the same ontology group simultaneously alter their expression pattern after induced epithelial damage, illustrating their complementary role during tissue regeneration. Conclusions An optimized semantic clustering protocol demonstrates the potential to reveal a readable and unique pattern in the expression profiling of GIT-specific Ub-ligases, exposing ontologically relevant gene clusters with potentially redundant features. This extends our knowledge of ontological relationships among Ub-ligases and ubiquitination-related genes, providing an alternative and more functional gene classification. In a similar way, semantic cluster analysis could be used for studding of other enzyme families, tissues and systems.

Air pollution caused by road traffic has an unfavorable impact on the environment and also on human health. It has previously been shown, that complete gasoline emissions lead to toxic effects in cell models originating from human airways. Here we focused on extractable organic matter (EOM) from particulate matter, collected from gasoline emissions from fuels with different ethanol content. We performed cytotoxicity evaluation, quantification of mucin and extracellular reactive oxygen species (ROS) production, DNA breaks detection, and selected gene deregulation analysis, after one and five days of exposure of human bronchial epithelial model (BEAS-2B) and a 3D model of the human airway (MucilAir (TM)). Our data suggest that the longer exposure had more pronounced effects on the parameters of cytotoxicity and mucin production, while the impacts on ROS generation and DNA integrity were limited. In both cell models the expression of CYP1A1 was induced, regardless of the exposure period or EOM tested. Several other genes, including FMO2, IL1A, or TNF, were deregulated depending on the exposure time. In conclusion, ethanol content in the fuels did not significantly impact the toxicity of EOM. Biological effects were mostly linked to xenobiotics metabolism and inflammatory response. BEAS-2B cells were more sensitive to the treatment.

Deciphering the role of the aryl hydrocarbon receptor (AhR) in lung cancer cells may help us to better understand the role of toxic AhR ligands in lung carcinogenesis, including cancer progression. We employed human lung carcinoma A549 cells to investigate their fate after continuous two-week exposure to model AhR agonists, genotoxic benzo[a]pyrene (BaP; 1 mu M) and non-genotoxic 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD; 10 nM). While TCDD increased proliferative rate of A549 cells, exposure to BaP decreased cell proliferation and induced epithelial-to-mesenchymal transition (EMT)-like phenotype, which was associated with enhanced cell migration, invasion, and altered cell morphology. Although TCDD also suppressed expression of E-cadherin and activated some genes linked to EMT, it did not induce the EMT-like phenotype. The results of transcriptomic analysis, and the opposite effects of BaP and TCDD on cell proliferation, indicated that a delay in cell cycle progression, together with a slight increase of senescence (when coupled with AhR activation), favors the induction of EMT-like phenotype. The shift towards EMT-like phenotype observed after simultaneous treatment with TCDD and mitomycin C (an inhibitor of cell proliferation) confirmed the hypothesis. Since BaP decreased cell proliferative rate via induction of p21 expression, we generated the A549 cell model with reduced p21 expression and exposed it to BaP for two weeks. The p21 knockdown suppressed the BaP-mediated EMT-like phenotype in A549 cells, thus confirming that a delayed cell cycle progression, together with p21-dependent induction of senescence-related chemokine CCL2, may contribute to induction of EMT-like cell phenotype in lung cells exposed to genotoxic AhR ligands.

Gene expression profiling was made more cost-effective by the NIH LINCS program that profiles only ∼1, 000 selected landmark genes and uses them to reconstruct the whole profile. The D–GEX method employs neural networks to infer the entire profile. However, the original D–GEX can be significantly improved. We propose a novel transformative adaptive activation function that improves the gene expression inference even further and which generalizes several existing adaptive activation functions. Our improved neural network achieves an average mean absolute error of 0.1340, which is a significant improvement over our reimplementation of the original D–GEX, which achieves an average mean absolute error of 0.1637. The proposed transformative adaptive function enables a significantly more accurate reconstruction of the full gene expression profiles with only a small increase in the complexity of the model and its training procedure compared to other methods.

Gasoline engine emissions have been classified as possibly carcinogenic to humans and represent a significant health risk. In this study, we used MucilAir (TM), a three-dimensional (3D) model of the human airway, and BEAS-2B, cells originating from the human bronchial epithelium, grown at the air-liquid interface to assess the toxicity of ordinary gasoline exhaust produced by a direct injection spark ignition engine. The transepithelial electrical resistance (TEER), production of mucin, and lactate dehydrogenase (LDH) and adenylate kinase (AK) activities were analyzed after one day and five days of exposure. The induction of double-stranded DNA breaks was measured by the detection of histone H2AX phosphorylation. Next-generation sequencing was used to analyze the modulation of expression of the relevant 370 genes. The exposure to gasoline emissions affected the integrity, as well as LDH and AK leakage in the 3D model, particularly after longer exposure periods. Mucin production was mostly decreased with the exception of longer BEAS-2B treatment, for which a significant increase was detected. DNA damage was detected after five days of exposure in the 3D model, but not in BEAS-2B cells. The expression of CYP1A1 and GSTA3 was modulated in MucilAir (TM) tissues after 5 days of treatment. In BEAS-2B cells, the expression of 39 mRNAs was affected after short exposure, most of them were upregulated. The five days of exposure modulated the expression of 11 genes in this cell line. In conclusion, the ordinary gasoline emissions induced a toxic response in MucilAir (TM). In BEAS-2B cells, the biological response was less pronounced, mostly limited to gene expression changes.

Preformed antibodies directed against donor human leukocyte antigen (HLA) antigens represent a major obstacle in kidney transplantation, limiting both access to transplantation and kidney allograft survival (1, 2). It is widely accepted that kidney transplantation across donor-specific antibodies (DSA) identified either by solid-phase assays or flow cytometry crossmatch (FCXM) is associated with a higher risk of antibody-mediated rejection (ABMR) and inferior allograft outcomes, even in absence of positive complement-dependent cytotoxicity crossmatch (complement-dependent cytotoxicity crossmatch [CDC XM]) (3–7). Several transplant programs have implemented peritransplant desensitization regimens using T- and B-cell depleting antibody induction, peritransplant apheresis, and high-dose intravenous immunoglobulin (IVIg) to counteract the deleterious effects of preformed DSA (8). Despite intense strategies of desensitization, there is still an increased rejection risk, which may critically depend on the strength of the preformed DSA. Previously, the Viennese group used anti-thymocyte globulin (ATG) induction and peritransplant immunoadsorption (IA) as desensitization regimens in DSA-positive deceased donor kidney transplantation. The only predictor of antibody-mediated rejection found by this study was donor-specific antibody mean fluorescence intensity (DSA MFI) (9). FCXM may have several advantages over DSA MFI in terms of better predictive power to select grafts at risk of ABMR (10, 11). Moreover, kidney transplantation with a low level of DSA with or without a low positive B-cell FCXM was found to be associated with satisfactory outcomes in highly sensitized mostly living donor kidney transplant recipients who received depleting antibody induction and frequently also desensitization (12).

Motor vehicle emissions substantially contribute to air pollution worldwide and cause serious health problems. While the deleterious effects of diesel exhaust particulate matter (PM) have been widely studied, much less attention is paid to toxicity of PM emitted by gasoline engines although they also produce considerable amount of PM. The primary objective of this research was to assess toxic potencies of exhaust PM released by conventional gasoline engine fueled with neat gasoline (E0) or gasoline-ethanol blend (15% ethanol, v/v, E15). Despite a similar particle mass (μg PM/kg fuel) produced by both fuels, PM emitted by E15 contained higher amount of harmful polycyclic aromatic hydrocarbons (PAH) as suggested by chemical analysis. To examine the toxicity of organic PM constituents, human lung BEAS-2B cells were exposed for 4h and 24h to a subtoxic dose of E0 and E15 PM organic extracts. We used genome scale transcriptomic analysis to characterize the toxic response and to identify modulated biological process and pathways. Whereas 4h exposure to both PM extracts resulted in modulation of similar genes and pathways related to lipid and steroid metabolism, activation of PPARa, oxidative stress and immune response, 24h exposure was more specific for each extract; although both induced expression of PAH-metabolic enzymes, modulated metabolism of lipids or activated PPARa, E15 additionally deregulated variety of other pathways. Overall, the PM mass produced by both fuels was similar, however, higher PAH content in E15 PM organic extract may have contributed to more extensive toxic response particularly after 24h exposure in BEAS-2B cells.

Emissions from road traffic are among the major contributors to air pollution worldwide and represent a serious environmental health risk. Although traffic-related pollution has been most commonly associated with diesel engines, increasing evidence suggests that gasoline engines also produce a considerable amount of potentially hazardous particulate matter (PM). The primary objective of this study was to compare the intrinsic toxic properties of the organic components of PM, generated by a conventional gasoline engine fueled with neat gasoline (E0), or gasoline-ethanol blend (15 % ethanol, v/v, E15). Our results showed that while E15 has produced, compared to gasoline and per kg of fuel, comparable particle mass (mu g PM/kg fuel) and slightly more particles by number, the organic extract from the particulate matter produced by E15 contained a larger amount of harmful polycyclic aromatic hydrocarbons (PAHs), as determined by the chemical analysis. To examine the toxicity, we monitored genome-wide gene expression changes in human lung BEAS-2B cells, exposed for 4 h and 24 h to a subtoxic dose of each PM extract. After 4 h exposure, numerous dysregulated genes and processes such as oxidative stress, lipid and steroid metabolism, PPAR alpha signaling and immune response, were found to be common for both extract treatments. On the other hand, 24 h exposure resulted in more distinctive gene expression patterns. Although we identified several common modulated processes indicating the metabolism of PAHs and activation of aryl hydrocarbon receptor (AhR), E15 specifically dysregulated a variety of other genes and pathways related to cancer promotion and progression. Overall, our findings suggest that the ethanol addition to gasoline changed the intrinsic properties of PM emissions and increased the PAH content in PM organic extract, thus contributing to a more extensive toxic response particularly after 24 h exposure in BEAS-2B cells.

Myelodysplastic syndromes (MDS) are hematopoietic stem cell disorders with large heterogeneity at the clinical and molecular levels. As diagnostic procedures shift from bone marrow biopsies towards less invasive techniques, circulating small noncoding RNAs (sncRNAs) have become of particular interest as potential novel noninvasive biomarkers of the disease. We aimed to characterize the expression profiles of circulating sncRNAs of MDS patients and to search for specific RNAs applicable as potential biomarkers. We performed small RNA-seq in paired samples of total plasma and plasma-derived extracellular vesicles (EVs) obtained from 42 patients and 17 healthy controls and analyzed the data with respect to the stage of the disease, patient survival, response to azacitidine, mutational status, and RNA editing. Significantly higher amounts of RNA material and a striking imbalance in RNA content between plasma and EVs (more than 400 significantly deregulated sncRNAs) were found in MDS patients compared to healthy controls. Moreover, the RNA content of EV cargo was more homogeneous than that of total plasma, and different RNAs were deregulated in these two types of material. Differential expression analyses identified that many hematopoiesis-related miRNAs (e.g., miR-34a, miR-125a, and miR-150) were significantly increased in MDS and that miRNAs clustered on 14q32 were specifically increased in early MDS.

The risk of exposure to nanoparticles (NPs) has rapidly increased during the last decade due to the vast use of nanomaterials (NMs) in many areas of human life. Despite this fact, human biomonitoring studies focused on the effect of NP exposure on DNA alterations are still rare. Furthermore, there are virtually no epigenetic data available. In this study, we investigated global and gene-specific DNA methylation profiles in a group of 20 long-term (mean 14.5 years) exposed, nanocomposite, research workers and in 20 controls. Both groups were sampled twice/day (pre-shift and post-shift) in September 2018. We applied Infinium Methylation Assay, using the Infinium MethylationEPIC BeadChips with more than 850,000 CpG loci, for identification of the DNA methylation pattern in the studied groups. Aerosol exposure monitoring, including two nanosized fractions, was also performed as proof of acute NP exposure. The obtained array data showed significant differences in methylation between the exposed and control groups related to long-term exposure, specifically 341 CpG loci were hypomethylated and 364 hypermethylated. The most significant CpG differences were mainly detected in genes involved in lipid metabolism, the immune system, lung functions, signaling pathways, cancer development and xenobiotic detoxification. In contrast, short-term acute NP exposure was not accompanied by DNA methylation changes. In summary, long-term (years) exposure to NP is associated with DNA epigenetic alterations.

Identification of non-trivial and meaningful patterns in omics data is one of the most important biological tasks. The patterns help to better understand biological systems and interpret experimental outcomes. A well-established method serving to explain such biological data is Gene Set Enrichment Analysis. However, this type of analysis is restricted to a specific type of evaluation. Abstracting from details, the analyst provides a sorted list of genes and ontological annotations of the individual genes; the method outputs a subset of ontological terms enriched in the gene list. Here, in contrary to enrichment analysis, we introduce a new tool/framework that allows for the induction of more complex patterns of 2-dimensional binary omics data. This extension allows to discover and describe semantically coherent biclusters.

We investigated the transcriptomic response and epigenetic changes in the lungs of mice exposed to inhalation of copper(II) oxide nanoparticles (CuO NPs) (8 x 10(5) NPs/m(3)) for periods of 3 days, 2 weeks, 6 weeks, and 3 months. A whole genome transcriptome and miRNA analysis was performed using next generation sequencing. Global DNA methylation was assessed by ELISA. The inhalation resulted in the deregulation of mRNA transcripts: we detected 170, 590, 534, and 1551 differentially expressed transcripts after 3 days, 2 weeks, 6 weeks, and 3 months of inhalation, respectively. Biological processes and pathways affected by inhalation, differed between 3 days exposure (collagen formation) and longer treatments (immune response). Periods of two weeks exposure further induced apoptotic processes, 6 weeks of inhalation affected the cell cycle, and 3 months of treatment impacted the processes related to cell adhesion. The expression of miRNA was not affected by 3 days of inhalation. Prolonged exposure periods modified miRNA levels, although the numbers were relatively low (17, 18, and 38 miRNAs, for periods of 2 weeks, 6 weeks, and 3 months, respectively). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways analysis based on miRNA-mRNA interactions, revealed the deregulation of processes implicated in the immune response and carcinogenesis. Global DNA methylation was not significantly affected in any of the exposure periods. In summary, the inhalation of CuO NPs impacted on both mRNA and miRNA expression. A significant transcriptomic response was already observed after 3 days of exposure. The affected biological processes and pathways indicated the negative impacts on the immune system and potential role in carcinogenesis.

Background: myelodysplastic syndrome (MDS) is a hematopoietic stem cell disorder with an incompletely known pathogenesis. Long noncoding RNAs (lncRNAs) play multiple roles in hematopoiesis and represent a new class of biomarkers and therapeutic targets, but information on their roles in MDS is limited. Aims: here, we aimed to characterize lncRNAs deregulated in MDS that may function in disease pathogenesis. In particular, we focused on the identification of lncRNAs that could serve as novel potential biomarkers of adverse outcomes in MDS. Methods: we performed microarray expression profiling of lncRNAs and protein-coding genes (PCGs) in the CD34+ bone marrow cells of MDS patients. Expression profiles were analyzed in relation to different aspects of the disease (i.e., diagnosis, disease subtypes, cytogenetic and mutational aberrations, and risk of progression). LncRNA-PCG networks were constructed to link deregulated lncRNAs with regulatory mechanisms associated with MDS. Results: we found several lncRNAs strongly associated with disease pathogenesis (e.g., H19, WT1-AS, TCL6, LEF1-AS1, EPB41L4A-AS1, PVT1, GAS5, and ZFAS1). Of these, downregulation of LEF1-AS1 and TCL6 and upregulation of H19 and WT1-AS were associated with adverse outcomes in MDS patients Multivariate analysis revealed that the predominant variables predictive of survival are blast count, H19 level, and TP53 mutation. Coexpression network data suggested that prognosis-related lncRNAs are predominantly related to cell adhesion and differentiation processes (H19 and WT1-AS) and mechanisms such as chromatin modification, cytokine response, and cell proliferation and death (LEF1-AS1 and TCL6).

The fate of transplanted kidneys is substantially influenced by graft quality, with transplantation of kidneys from elderly and expanded criteria donors (ECDs) associated with higher occurrence of delayed graft function, rejection, and inferior long-term outcomes. However, little is known about early molecular fingerprints of these events in different donor categories. Borderline changes represent the most frequent histological finding early after kidney transplantation. Therefore, we examined outcomes and transcriptomic profiles of early-case biopsies diagnosed as borderline changes in different donor categories. In this single-center, retrospective, observational study, we compared midterm outcomes of kidney transplant recipients with early borderline changes as a first pathology between ECD (n = 109), standard criteria donor (SCDs, n = 109), and living donor (LD, n = 51) cohorts. Intragraft gene expression profiling by microarray was performed in part of these ECD, SCD, and LD cohorts. Although 5 year graft survival in patients with borderline changes in early-case biopsies was not influenced by donor category (log-rank P = 0.293), impaired kidney graft function (estimated glomerular filtration rate by Chronic Kidney Disease Epidemiology Collaboration equation) at M3, 1, 2, and 3 years was observed in the ECD cohort (P < 0.001). Graft biopsies from ECD donors had higher vascular intimal fibrosis and arteriolar hyalinosis compared to SCD and LD (P < 0.001), suggesting chronic vascular changes. Increased transcripts typical for ECD, as compared to both LD and SCD, showed enrichment of the inflammatory, defense, and wounding responses and the ECM-receptor interaction pathway. Additionally, increased transcripts in ECD vs. LD showed activation of complement and coagulation and cytokine-cytokine receptor pathways along with platelet activation and cell cycle regulation.

The Banff 2019 kidney allograft pathology update excluded isolated tubulitis without interstitial inflammation (ISO-T) from the category of borderline (suspicious) for acute T cell-mediated rejection due to its proposed benign clinical outcome. In this study, we explored the molecular assessment of ISO-T. ISO-T or interstitial inflammation with tubulitis (I+T) was diagnosed in indication biopsies within the first 14 postoperative days. The molecular phenotype of ISO-T was compared to I+T either by using RNA sequencing (n=16) or by Molecular Microscope Diagnostic System (MMDx, n=51). RNA sequencing showed lower expression of genes related to interferon-y (p=1.5 *10(-16)), cytokine signaling (p=2.1 *10(-20)) and inflammatory response (p=1.0*10(-13)) in the ISO-T group than in I+T group. Transcripts with increased expression in the I+T group overlapped significantly with previously described pathogenesis-based transcript sets associated with cytotoxic and effector T cell transcripts, and with T cell-mediated rejection (TCMR). MMDx classified 25/32 (78%) ISO-T biopsies and 12/19 (63%) I+T biopsies as no-rejection. ISO-T had significantly lower MMDx scores for interstitial inflammation (p=0.014), tubulitis (p=0.035) and TCMR (p=0.016) compared to I+T. Fewer molecular signals of inflammation in isolated tubulitis suggest that this is also a benign phenotype on a molecular level.

Background: One possible approach how to economically facilitate gene expression profiling is to use the L1000 platform which measures the expression of ~1,000 landmark genes and uses a computational method to infer the expression of another ~10,000 genes. One such method for the gene expression inference is a D-GEX which employs neural networks. Results: We propose two novel D-GEX architectures that significantly improve the quality of the inference by increasing the capacity of a network without any increase in the number of trained parameters. The architectures partition the network into individual towers. Our best proposed architecture - a checkerboard architecture with a skip connection and five towers - together with minor changes in the training protocol improves the average mean absolute error of the inference from 0.134 to 0.128. Conclusions: Our proposed approach increases the gene expression inference accuracy without increasing the number of weights of the model and thus without increasing the memory footprint of the model that is limiting its usage.

This study presents a toxicological evaluation of two types of carbon dots (CD), similar in size (<10 nm) but differing in surface charge. Whole-genome mRNA and miRNA expression (RNAseq), as well as gene-specific DNA methylation changes, were analyzed in human embryonic lung fibroblasts (HEL 12469) after 4 h and 24 h exposure to concentrations of 10 and 50 mu g/mL (for positive charged CD; pCD) or 10 and 100 mu g/mL (for negative charged CD, nCD). The results showed a distinct response for the tested nanomaterials (NMs). The exposure to pCD induced the expression of a substantially lower number of mRNAs than those to nCD, with few commonly differentially expressed genes between the two CDs. For both CDs, the number of deregulated mRNAs increased with the dose and exposure time. The pathway analysis revealed a deregulation of processes associated with immune response, tumorigenesis and cell cycle regulation, after exposure to pCD. For nCD treatment, pathways relating to cell proliferation, apoptosis, oxidative stress, gene expression, and cycle regulation were detected. The expression of miRNAs followed a similar pattern: more pronounced changes after nCD exposure and few commonly differentially expressed miRNAs between the two CDs. For both CDs the pathway analysis based on miRNA-mRNA interactions, showed a deregulation of cancer-related pathways, immune processes and processes involved in extracellular matrix interactions. DNA methylation was not affected by exposure to any of the two CDs. In summary, although the tested CDs induced distinct responses on the level of mRNA and miRNA expression, pathway analyses revealed a potential common biological impact of both NMs independent of their surface charge.

The aryl hydrocarbon receptor (AhR) activation has been shown to alter proliferation, apoptosis, or differentiation of adult rat liver progenitors. Here, we investigated the impact of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)-mediated AhR activation on a human model of bipotent liver progenitors, undifferentiated HepaRG cells. We used both intact undifferentiated HepaRG cells, and the cells with silenced Hippo pathway effectors, yes-associated protein 1 (YAP) and transcriptional coactivator with PDZ-binding motif (TAZ), which play key role(s) in tissue-specific progenitor cell self-renewal and expansion, such as in liver, cardiac, or respiratory progenitors. TCDD induced cell proliferation in confluent undifferentiated HepaRG cells; however, following YAP, and, in particular, double YAP/TAZ knockdown, TCDD promoted induction of apoptosis. These results suggested that, unlike in mature hepatocytes, or hepatocyte-like cells, activation of the AhR may sensitize undifferentiated HepaRG cells to apoptotic stimuli. Induction of apoptosis in cells with silenced YAP/TAZ was associated with upregulation of death ligand TRAIL, and seemed to involve both extrinsic and mitochondrial apoptosis pathways. Global gene expression analysis further suggested that TCDD significantly altered expression of constituents and/or transcriptional targets of signaling pathways participating in control of expansion or differentiation of liver progenitors, including EGFR, Wnt/beta-catenin, or tumor growth factor-beta signaling pathways. TCDD significantly upregulated cytosolic proapoptotic protein BMF (Bcl-2 modifying factor) in HepaRG cells, which could be linked with an enhanced sensitivity of TCDD-treated cells to apoptosis. Our results suggest that, in addition to promotion of cell proliferation and alteration of signaling pathways controlling expansion of human adult liver progenitors, AhR ligands may also sensitize human liver progenitor cells to apoptosis.

Objectives This study aimed to assess the biological impact of occupational exposure to diesel exhaust (DE) including DE particles (DEP) from heavy-duty diesel-powered equipment in Norwegian tunnel finishing workers (TFW).Methods TFW (n=69) and referents (n=69) were investigated for bulky DNA adducts (by 32P-postlabelling) and expression of microRNAs (miRNAs) (by small RNA sequencing) in peripheral blood mononuclear cells (PBMC), as well as circulating free arachidonic acid (AA) and eicosanoid profiles in plasma (by liquid chromatography–tandem mass spectrometry).Results PBMC from TFW showed significantly higher levels of DNA adducts compared with referents. Levels of DNA adducts were also related to smoking habits. Seventeen miRNAs were significantly deregulated in TFW. Several of these miRNAs are related to carcinogenesis, apoptosis and antioxidant effects. Analysis of putative miRNA-gene targets revealed deregulation of pathways associated with cancer, alterations in lipid molecules, steroid biosynthesis and cell cycle. Plasma profiles showed higher levels of free AA and 15-hydroxyeicosatetraenoic acid, and lower levels of prostaglandin D2 and 9-hydroxyoctadecadienoic acid in TFW compared with referents.Conclusion Occupational exposure to DE/DEP is associated with biological alterations in TFW potentially affecting lung homoeostasis, carcinogenesis, inflammation status and the cardiovascular system. Of particular importance is the finding that tunnel finishing work is associated with an increased level of DNA adducts formation in PBMC.

As severe dropout in single-cell RNA sequencing (scRNA-seq) degrades data quality, current methods for network in-ference face increased uncertainty from such data. To exa-mine how dropout influences directional dependency infe-rence from scRNA-seq data, we thus studied four methodsbased on discrete data that are model-free without paramet-ric model assumptions. They include two established me-thods: conditional entropy and Kruskal-Wallis test, and tworecent methods: causal inference by stochastic complexityand function index. We also included three non-directionalmethods for a contrast. On simulated data, function indexperformed most favorably at varying dropout rates, samplesizes, and discrete levels. On an scRNA-seq dataset from de-veloping mouse cerebella, function index and Kruskal-Wallistest performed favorably over other methods in detectingexpression of developmental genes as a function of time.Overall among the four methods, function index is mostresistant to dropout for both directional and dependencyinference. The next best choice, Kruskal-Wallis test, carriesa directional bias towards a uniformly distributed variable.We conclude that a method robust to marginal distributi-ons with a sufficiently large sample size can reap benefitsof single-cell over bulk RNA sequencing in understandingmolecular mechanisms at the cellular resolution.

In the body, engineered nanoparticles (NPs) may be recognized and processed by immune cells, among which macrophages play a crucial role. We evaluated the effects of selected NPs [NM-100 (TiO2), NM-110 (ZnO), NM-200 (SiO2), and NM-300 K (Ag)] on THP-1 macrophage-like cells. The cells were exposed to subcytotoxic concentrations of NPs (1–25 µg/mL) and the expression of immunologically relevant genes (VCAM1, TNFA, CXCL8, ICAM1, CD86, CD192, and IL1B) was analyzed by RT-qPCR. The expression of selected cytokines, growth factors and surface molecules was assessed by flow cytometry or ELISA. Generation of reactive oxygen species and induction of DNA breaks were also analyzed. Exposure to diverse NPs caused substantially different molecular responses. No significant effects were detected for NM-100 treatment. NM-200 induced production of IL-8, a potent attractor and activator of neutrophils, growth factors (VEGF and IGF-1) and superoxide. NM-110 triggered a proinflammatory response, characterized by the activation of transcription factor NF-κB, an enhanced production of proinflammatory cytokines (TNF-α) and chemokines (IL-8). Furthermore, the expression of cell adhesion molecules VCAM-1 and ICAM-1 and hepatocyte growth factor (HGF), as well as superoxide production and DNA breaks, were affected. NM-300 K enhanced IL-8 production and induced DNA breaks, however, it decreased the expression of chemokine receptor (CCR2) and CD86 molecule, indicating potential immunosuppressive activity. The toxicity of ZnO and Ag NPs was probably caused by their intracellular dissolution, as indicated by transmission electron microscopy imaging. The observed effects in macrophages might further influence both innate and adaptive immune responses by promoting neutrophil recruitment via IL-8 release and enhancing the adhesion and stimulation of T cells by VCAM-1 and ICAM-1 expression.

TiO2 along with nano-TiO2 are commonly found in consumer products. In vivo studies have observed an accumulation of nano-TiO2 in macrophages. However, characteristics of nano-TiO2 determining toxicity remain unclear. In our study, the cytotoxic effects of 14 diverse nano-TiO2 on THP-1 macrophage-like cells were measured by 3 cytotoxicity assays (MTS, WST-1 and LDH). Total averaged cytotoxicity was calculated using principal component analysis. Characteristics of all 14 nano-TiO2 included hydrodynamic diameter, zeta potential, shape, polydispersity index (PDI) and concentration; moreover, crystal form, specific surface area and crystallite size were measured for 10 nano-TiO2.The variables affecting cytotoxicity were chosen using LASSO (least absolute shrinkage and selection operator). Except for concentration, PDI in media measured within 1 h after preparation of the nanomaterial dispersion was selected as a variable affecting cytotoxicity: stable dispersion resulted in higher cytotoxic effects. Crystallite size has been shown to have nonlinear effects (particles of sizes between 20 and 60 nm were cytotoxic while smaller and larger ones were not) and thus it has been excluded from LASSO. The shape (particles/fibre) and crystal form did not affect the cytotoxicity. PDI and the nonlinear effect of size could be an explanation for the inconsistencies of the cytotoxicity of nano-TiO2 in various studies.

The biological effects induced by complete engine emissions in a 3D model of the human airway (MucilAirTM) and in human bronchial epithelial cells (BEAS-2B) grown at the air-liquid interface were compared. The cells were exposed for one or five days to emissions generated by a Euro 5 direct injection spark ignition engine. The general condition of the cells was assessed by the measurement of transepithelial electrical resistance and mucin production. The cytotoxic effects were evaluated by adenylate kinase (AK) and lactate dehydrogenase (LDH) activity. Phosphorylation of histone H2AX was used to detect double-stranded DNA breaks. The expression of the selected 370 relevant genes was analyzed using next-generation sequencing. The exposure had minimal effects on integrity and AK leakage in both cell models. LDH activity and mucin production in BEAS-2B cells significantly increased after longer exposures; DNA breaks were also detected. The exposure affected CYP1A1 and HSPA5 expression in MucilAirTM. There were no effects of this kind observed in BEAS-2B cells; in this system gene expression was rather affected by the time of treatment. The type of cell model was the most important factor modulating gene expression. In summary, the biological effects of complete emissions exposure were weak. In the specific conditions used in this study, the effects observed in BEAS-2B cells were induced by the exposure protocol rather than by emissions and thus this cell line seems to be less suitable for analyses of longer treatment than the 3D model.

Exposure to persistent ligands of aryl hydrocarbon receptor (AhR) has been found to cause lung cancer in experimental animals, and lung adenocarcinomas are often associated with enhanced AhR expression and aberrant AhR activation. In order to better understand the action of toxic AhR ligands in lung epithelial cells, we performed global gene expression profiling and analyze TCDD-induced changes in A549 transcriptome, both sensitive and non-sensitive to CH223191 co-treatment. Comparison of our data with results from previously reported microarray and ChIP-seq experiments enabled us to identify candidate genes, which expression status reflects exposure of lung cancer cells to TCDD, and to predict processes, pathways (e.g. ER stress, Wnt/β-cat, IFNɣ, EGFR/Erbb1), putative TFs (e.g. STAT, AP1, E2F1, TCF4), which may be implicated in adaptive response of lung cells to TCDD-induced AhR activation. Importantly, TCDD-like expression fingerprint of selected genes was observed also in A549 cells exposed acutely to both toxic (benzo[a]pyrene, benzo[k]fluoranthene) and endogenous AhR ligands (2-(1H-Indol-3-ylcarbonyl)-4-thiazolecarboxylic acid methyl ester and 6-formylindolo[3,2-b]carbazole). Overall, our results suggest novel cellular candidates, which could help to improve monitoring of AhR-dependent transcriptional activity during acute exposure of lung cells to distinct types of environmental pollutants.

Intimal arteritis is known to be a negative prognostic factor for kidney allograft survival. Isolated v-lesion (IV) is defined as intimal arteritis with minimal tubulointerstitial inflammation (TI). Although the Banff classification assesses IV as T cell-mediated rejection (TCMR), clinical, and prognostic significance of early IV (early IV, eIV) with negative C4d and donor-specific antibodies (DSA) remains unclear. To help resolve if such eIV truly represents acute rejection, a molecular study was performed. The transcriptome of eIV (n=6), T cell-mediated vascular rejection with rich TI (T cell-mediated vascular rejection, TCMRV, n=4) and non-rejection histologic findings (n=8) was compared using microarrays. A total of 310 genes were identified to be deregulated in TCMRV compared with eIV. Gene enrichment analysis categorized deregulated genes to be associated primarily with T-cells associated biological processes involved in an innate and adaptive immune and inflammatory response. Comparison of deregulated gene lists between the study groups and controls showed only a 1.7% gene overlap. Unsupervised hierarchical cluster analysis revealed clear distinction of eIV from TCMRV and showed similarity with a control group. Up-regulation of immune response genes in TCMRV was validated using RT-qPCR in a different set of eIV (n=12) and TCMRV (n=8) samples. The transcriptome of early IV (< 1 month) with negative C4d and DSA is associated with a weak immune signature compared with TCMRV and shows similarity with normal findings. Such eIV may feature non-rejection origin and reflect an injury distinct from an alloimmune response. The present study supports use of molecular methods when interpreting kidney allograft biopsy findings.

Prenatal exposure to air pollution is associated with intrauterine growth restriction and low birth weight. Gene expression changes in newborns in relation to air pollution have not been sufficiently studied. We analyzed whole genome expression in cord blood leukocytes of 202 newborns from diverse localities of the Czech Republic, differing among other factors in levels of air pollution: the district of Karvina (characterized by higher concentration of air pollutants) and Ceske Budejovice (lower air pollution levels). We aimed to identify differentially expressed genes (DEGs) and pathways in relation to locality and concentration of air pollutants. We applied the linear model to identify the specific DEGs and the correlation analysis, to investigate the relationship between the concentrations of air pollutants and gene expression data. An analysis of biochemical pathways and gene set enrichment was also performed. In general, we observed modest changes of gene expression, mostly attributed to the effect of the locality. The highest number of DEGs was found in samples from the district of Karvina. A pathway analysis revealed a deregulation of processes associated with cell growth, apoptosis or cellular homeostasis, immune response‐related processes or oxidative stress response. The association between concentrations of air pollutants and gene expression changes was weak, particularly for samples collected in Karvina. In summary, as we did not find a direct effect of exposure to air pollutants, we assume that the general differences in the environment, rather than actual concentrations of individual pollutants, represent a key factor affecting gene expression changes at delivery.

Rule learning is a kind of machine learning method that induces a set of classification rules from a given set of training examples. As a well-known representative of this learners, we can adduce CN2, RIPPER, or PRIM. All of them use if-then statement for corresponding hypothesis formulation where the antecedent is in the form of a conjunction of logical terms, and the consequent is a class label. From a bioinformatician point of view, these learners are suitable especially for their easy and clear interpretation of hypothesis on the contrary of a neural network, for example. The other thing that can help biologists interpret their data in a more natural way is a background knowledge. Nowadays, the most popular form of background knowledge in the field of bioinformatics are ontologies, especially Gene Ontology or Disease Ontology. There are other types of structured databases such as KEGG, that can also be interpreted as an ontology or a taxonomy. In our work, we combine these two concepts, rule learning and ontologies/taxonomies, together

Modern vehicles equipped with Gasoline Direct Injection (GDI) engine have emerged as an important source of particulate emissions potentially harmful to human health. We collected and characterized gasoline exhaust particles (GEPs) produced by neat gasoline fuel (E0) and its blends with 15% ethanol (E15), 25% n-butanol (n-But25) and 25% isobutanol (i-But25). To study the toxic effects of organic compounds extracted from GEPs, we analyzed gene expression profiles in human lung BEAS-2B cells. Despite the lowest GEP mass, n-But25 extract contained the highest concentration of polycyclic aromatic hydrocarbons (PAHs), while i-But25 extract the lowest. Gene expression analysis identified activation of the DNA damage response and other subsequent events (cell cycle arrest, modulation of extracellular matrix, cell adhesion, inhibition of cholesterol biosynthesis) following 4 h exposure to all GEP extracts. The i-But25 extract induced the most distinctive gene expression pattern particularly after 24 h exposure. Whereas E0, E15 and n-But25 extract treatments resulted in persistent stress signaling including DNA damage response, MAPK signaling, oxidative stress, metabolism of PAHs or pro-inflammatory response, i-But25 induced changes related to the metabolism of the cellular nutrients required for cell recovery. Our results indicate that i-But25 extract possessed the weakest genotoxic potency possibly due to the low PAH content.

From the perspective of the immune system, nanomaterials (NMs) represent invading agents. Macrophages are immune cells residing in all organs and tissues as the first line of defense. Interactions of macrophages with NMs can determine the fate of NMs as well as their potential toxic effects. In the present study, we compared toxicity of four different types of NMs [NM-100 (TiO2, 110 nm), NM-110 (ZnO, 20 nm), NM-200 (SiO2, 150 nm) and NM-300K (Ag, 20 nm)], towards THP-1 macrophage-like cells. Cells were incubated with non-cytotoxic concentrations (1-25 µg/ml) of NMs for 24 hours and microarray technology was used to analyze changes in whole-genome expression. Gene expression profiling revealed a substantially different molecular response following exposure to diverse NMs. While NM-100 did not exert any significant effect on gene expression profile, all other NMs triggered a pro-inflammatory response characterized by an activation of the NF-κB transcription factor and induced expression of numerous chemokines and cytokines. NM-110 and NM-300K further modulated processes such as DNA damage response, oxidative and replication stress as well as cell cycle progression and proteasome function. We suppose that genotoxicity of ZnO and Ag NMs leading to DNA damage and alternatively to apoptosis in THP-1 macrophages is probably caused by the extensive intracellular dissolution of these NPs, as confirmed by TEM imaging.

To model the quantitative relationship of the nanoparticle toxicity we can use theoretical molecular descriptors or physico-chemical characteristics. The former provide an auspicious interpretation of the toxicity mechanisms, however their computation may be very demanding, namely in the nanoscale. The latter are on the other hand fully observable, yet scarcely available for all the toxicity-assessed particles. Currently, there are large initiatives generating data for QNTR, including their toxicity and physico-chemical features. Resulting data are naturally very heterogeneous because of multiple subjects involved in the project. In this study, we investigate whether the data generated from such large projects are sufficient to induce well-generalizing models. We used the data generated by MODENA-COST, consisting of the toxicity measurements and physico-chemical characteristics of 192 nanoparticles. We build several machine-learning based models and focused on their statistical validity. The internal evaluation of these models (i.e. protocol using the same data set, such as cross-validation) suggests quite good validity of these models. Then we employed a rigorous validation protocol and external data set of our own measurements related to 10 standardized MeOx nanoparticles. Hence, the result were not so optimistic at all. Instead, they seem valid only for a well-defined set of experimental conditions.

Keratoconus is a relatively frequent disease leading to severe visual impairment. Existing therapies are imperfect and clinical management may benefit from improved understanding of mechanisms leading to this disease. We aim to investigate the replication of 11 single nucleotide polymorphisms (SNPs) with keratoconus.

Background: One of the major challenges in the analysis of gene expression data is to identify local patterns composed of genes showing coherent expression across subsets of experimental conditions. Such patterns may provide an understanding of underlying biological processes related to these conditions. This understanding can further be improved by providing concise characterizations of the genes and situations delimiting the pattern. Results: We propose a method called semantic biclustering with the aim to detect interpretable rectangular patterns in binary data matrices. As usual in biclustering, we seek homogeneous submatrices, however, we also require that the included elements can be jointly described in terms of semantic annotations pertaining to both rows (genes) and columns (samples). To find such interpretable biclusters, we explore two strategies. The first endows an existing biclustering algorithm with the semantic ingredients. The other is based on rule and tree learning known from machine learning. Conclusions: The two alternatives are tested in experiments with two Drosophila melanogaster gene expression datasets. Both strategies are shown to detect sets of compact biclusters with semantic descriptions that also remain largely valid for unseen (testing) data. This desirable generalization aspect is more emphasized in the strategy stemming from conventional biclustering although this is traded off by the complexity of the descriptions (number of ontology terms employed), which, on the other hand, is lower for the alternative strategy.

The need of creating user profiles arises from applications like Customer Relationship Management and Web Personalization. These customer models than can be used to create machine learning models that predict user’s preference in web pages or the likelihood of him buying a product. We developed a general method that can create user profiles, that capture long term as well as short term user behaviour. By using different types of windowed features we were able to create models in the Attribute Value Learning paradigm. These models perform well in the areas of gaming as well as large news site.

This study used toxicogenomics to identify the complex biological response of human lung BEAS-2B cells treated with organic components of particulate matter in the exhaust of a diesel engine. First, we characterized particles from standard diesel (B0), biodiesel (methylesters of rapeseed oil) in its neat form (B100) and 30% by volume blend with diesel fuel (B30), and neat hydrotreated vegetable oil (NEXBTL100). The concentration of polycyclic aromatic hydrocarbons (PAHs) and their derivatives in organic extracts was the lowest for NEXBTL100 and higher for biodiesel. We further analyzed global gene expression changes in BEAS-2B cells following 4 h and 24 h treatment with extracts. The concentrations of 50 µg extract/mL induced a similar molecular response. The common processes induced after 4 h treatment included antioxidant defense, metabolism of xenobiotics and lipids, suppression of pro-apoptotic stimuli, or induction of plasminogen activating cascade; 24 h treatment affected fewer processes, particularly those involved in detoxification of xenobiotics, including PAHs. The majority of distinctively deregulated genes detected after both 4 h and 24 h treatment were induced by NEXBTL100; the deregulated genes included, e.g., those involved in antioxidant defense and cell cycle regulation and proliferation. B100 extract, with the highest PAH concentrations, additionally affected several cell cycle regulatory genes and p38 signaling.

Molecular biology is a domain endowed by a good amount of data and well-formalized knowledge. Based on measured data and domain knowledge, an intelligent integrative analysis is capable of extracting new and more specific knowledge, which may help to comprehension of e.g. disease mechanisms. We have proposed miXGENE, a web service for integrating and analyzing high-throughput omics data, namely from the microarray-based expression or methylation measurements, together with formal biological knowledge, such as gene ontologies and curated or predicted omics interactions. The tool enables building the most employed analytical workflows for processing user-data or the data from public databases. Processing of the data is followed by their integrative statistical or machine-learning based analysis, and completed with the presentation of results in the expert-comprehensible terms. We propose an innovation of the tool which profits from the infrastructure of the Czech National Grid, CESNET – MetaCentrum, which facilitates the most computationally demanding sections.

We motivate and define the task of semantic biclustering. In an input gene expression matrix, the task is to discover homogeneous biclusters allowing joint characterization of the contained elements in terms of knowledge pertaining to both the rows (e.g. genes) and the columns (e.g. situations). We propose two approaches to solve the task, based on adaptations of current biclustering, enrichment, and rule and tree learning methods. We compare the approaches in experiments with Drosophila ovary gene expression data. Our findings indicate that both the proposed methods induce compact bicluster sets whose description is applicable to unseen data. The bicluster enrichment method achieves the best performance in terms of the area under the ROC curve, at the price of employing a large number of ontology terms to describe the biclusters.

Dvojshluková analýza dat genové exprese je používaná bioinformatická metoda pro vyhledávání význačných lokálních vzorů v genové matici. Matice genových expresí je uvedenou metodou analyzována, na rozdíl od klasických jednodimenzionálních shlukovacích metod, současně ve dvou dimenzích: v dimenzi obsahující geny (řádky) a v dimenzi obsahující jednotlivé vzorky (sloupce). Algoritmus pro dvojshlukování tedy identifikuje podskupiny genů v podskupině vzorků. Typickým příkladem sémantického shlukování při analýze genových dat je shlukování genů na základě hodnoty exprese, kde jednotlivé shluky mohou být charakterizovány pomocí termů z Gene ontology.

Azacitidine (AZA) is a hypomethylating drug used to treat disorders associated with myelodysplasia and related neoplasms. Approximately 50 % of patients do not respond to AZA and have very poor outcomes. There is thus great interest in identifying predictive biomarkers for AZA responsiveness. We searched for specific genes whose expression level was associated with response status. Using microarrays, we analyzed gene expression patterns in bone marrow CD34+ cells in serial samples from 32 patients with myelodysplastic syndromes, chronic myelomonocytic leukemia, and acute myeloid leukemia with myelodysplasia-related changes before and during the AZA therapy. At baseline, a comparison of the responders and non-responders showed 52 differentially expressed genes (P < 0.01). Functional annotation of the deregulated genes revealed categories primarily related to ribosomes and pathways associated with proliferation. The expression level of RPL28 correlated with overall survival. We identified altered expression in 167 genes in responders, 26 genes in non-responders with stable disease, and 13 genes in non-responders with disease progression using paired t test of expression levels in patients before and during treatment. Our data indicate that AZA treatment failure is associated with the up-regulation of ribosomal genes/pathways that are likely related to intensive proteosynthesis in proliferative/neoplastic cells of non-responders.

BACKGROUND: Lenalidomide is a novel thalidomide analog with immunomodulatory and antiangiogenic effects that has been successfully used for the treatment of low- and intermediate-1 risk myelodysplastic syndromes (MDS) with a del(5q) aberration. METHODS: Because information about the influence of lenalidomide on the miRNA transcriptome is limited, we performed miRNA expression profiling of bone marrow CD34+ cells obtained from MDS patients with del(5q) abnormality who had been subjected to lenalidomide treatment. To define differences in miRNA expression, we performed paired data analysis to compare the miRNA profiles of patients before and during lenalidomide treatment and those of healthy donors. RESULTS: The analysis showed that miRNAs clustering to the 14q32 region had a higher expression level in patient samples before treatment than in the healthy control samples, and this elevated expression was diminished following lenalidomide administration. CONCLUSION: Because some of the 14q32 miRNAs play important roles in hematopoiesis, stem cell differentiation and apoptosis induction, the expression of this cluster may be associated with the pathophysiology of the disease.

Lenalidomide is a potent drug with pleiotropic effects in patients with myelodysplastic syndrome (MDS) with deletion of the long arm of chromosome 5 [del(5q)]. We investigated its effect on regulation of microRNA (miRNA) expression profiles in del(5q) MDS patients in vivo.

The common problems in machine learning from omics data are the scarcity of samples, the high number of features and their complex interaction structure. The models built solely from measured data often suffer from overfitting. One of possible methods dealing with overfitting is to use prior knowledge for regularization. This work analyzes contribution of feature interaction networks in regularization of ensemble classifiers representing another approach to overfitting reduction. We study how utilization of feature interaction networks influences diversity of weak classifiers and thus accuracy of the resulting ensemble model. The network and its random walks are used to control the feature randomization during construction of weak classifiers, which makes them more diverse than in the well-known random forest. We experiment with different types of weak classifiers (trees, logistic regression, naive Bayes) and different random walk lengths and demonstrate that diversity of weak classifiers grows with increasing network locality of weak classifiers.

In this work we integrate conventional mRNA expression profiles with miRNA expressions using the knowledge of their validated or predicted interactions in order to improve class prediction in genetically determined diseases. The raw mRNA and miRNA expression features become enriched or replaced by new aggregated features that model the mRNA-miRNA interaction. The proposed subtractive integration method is directly motivated by the inhibition/degradation models of gene expression regulation. The method aggregates mRNA and miRNA expressions by subtracting a proportion of miRNA expression values from their respective target mRNAs. Further, its modification based on singular value decomposition that enables different subtractive weights for different miRNAs is introduced. Both the methods are used to model the outcome or development of myelodysplastic syndrome, a blood cell production disease often progressing to leukemia. The reached results demonstrate that the integration improves classification performance when dealing with mRNA and miRNA features of comparable significance. The proposed methods are available as a part of the web tool miXGENE.

Contemporary molecular biology deals with wide and heterogeneous sets of measurements to model and understand underlying biological processes including complex diseases. Machine learning provides a frequent approach to build such models. However, the models built solely from measured data often suffer from overfitting, as the sample size is typically much smaller than the number of measured features. In this paper, we propose a random forest-based classifier that reduces this overfitting with the aid of prior knowledge in the form of a feature interaction network. We illustrate the proposed method in the task of disease classification based on measured mRNA and miRNA profiles complemented by the interaction network composed of the miRNA–mRNA target relations and mRNA–mRNA interactions corresponding to the interactions between their encoded proteins. We demonstrate that the proposed network-constrained forest employs prior knowledge to increase learning bias and consequently to improve classification accuracy, stability and comprehensibility of the resulting model. The experiments are carried out in the domain of myelodysplastic syndrome that we are concerned about in the long term. We validate our approach in the public domain of ovarian carcinoma, with the same data form. We believe that the idea of a network-constrained forest can straightforwardly be generalized towards arbitrary omics data with an available and non-trivial feature interaction network. The proposed method is publicly available in terms of miXGENE system (http://mixgene.felk.cvut.cz), the workflow that implements the myelodysplastic syndrome experiments is presented as a dedicated case study.

BACKGROUND: Discovering gene interactions and their characterizations from biological text collections is a crucial issue in bioinformatics. Indeed, text collections are large and it is very difficult for biologists to fully take benefit from this amount of knowledge. Natural Language Processing (NLP) methods have been applied to extract background knowledge from biomedical texts. Some of existing NLP approaches are based on handcrafted rules and thus are time consuming and often devoted to a specific corpus. Machine learning based NLP methods, give good results but generate outcomes that are not really understandable by a user. RESULTS: We take advantage of an hybridization of data mining and natural language processing to propose an original symbolic method to automatically produce patterns conveying gene interactions and their characterizations. Therefore, our method not only allows gene interactions but also semantics information on the extracted interactions (e.g., modalities, biological contexts, interaction types) to be detected. Only limited resource is required: the text collection that is used as a training corpus. Our approach gives results comparable to the results given by state-of-the-art methods and is even better for the gene interaction detection in AIMed. CONCLUSIONS: Experiments show how our approach enables to discover interactions and their characterizations. To the best of our knowledge, there is few methods that automatically extract the interactions and also associated semantics information. The extracted gene interactions from PubMed are available through a simple web interface at https://bingotexte.greyc.fr/. The software is available at https://bingo2.greyc.fr/?q=node/22.

Current high-throughput technologies lead to the boost of omics data with thousands of features measured in parallel. The phenotype specific markers are learned from the data to better understand the disease mechanism and to build predictive models. However, the learning is prone to overfitting, caused by a small sample size and large feature space dimension. Consequently, resulting models are inaccurate and difficult to interpret due to the complex nature of omics processes. In this paper, we propose a methodology for learning simple yet biologically meaningful linear classification models. A linear support vector machine is trained; the learning is regularized by prior knowledge. Regularization parameters enable the expert to operatively adjust the interpretation of the models and their conformity with recent domain research while maintaining their accuracy. We performed robust experiments showing empirical validity of our methodology. In the study related to myelodysplastic syndrome we demonstrate the performance and interpretation of disease classification models. These models are consistent with recent progress in myelodysplastic syndrome research.

A number of toxic effects of respirable ambient air particles (genotoxic effects, inflammation, oxidative damage) have been attributed to organic compounds bound onto the particle surface. In this study, we analyzed global gene expression changes caused by the extractable organic matters (EOMs) from respirable airborne particles <2.5 μm (PM2.5), collected at 3 localities from heavily polluted areas of the Czech Republic and a control locality with low pollution levels, in human lung epithelial A549 cells. Although the sampled localities differed in both extent and sources of air pollution, EOMs did not induce substantially different gene expression profiles. The number of transcripts deregulated in A549 cells treated with the lowest EOM concentration (10 μg/ml) ranged from 65 to 85 in 4 sampling localities compared to the number of transcripts deregulated after 30 μg/ml and 60 μg/ml of EOMs, which ranged from 90 to 109, and from 149 to 452, respectively. We found numerous commonly deregulated genes and pathways related to activation of the aryl hydrocarbon receptor (AhR) and metabolism of xenobiotics and endogenous compounds. We further identified deregulation of expression of the genes involved in pro-inflammatory processes, oxidative stress response and in cancer and developmental pathways, such as TGF-β and Wnt signaling pathways. No cell cycle arrest, DNA repair or pro-apoptotic responses were identified at the transcriptional level after the treatment of A549 cells with EOMs. In conclusion, numerous processes and pathways deregulated in response to EOMs suggest a significant role of activated AhR. Interestingly, we did not observe substantial gene expression changes related to DNA damage response, possibly due to the antagonistic effect of non-genotoxic EOM components.

In this study, we compared the genotoxicity and aryl hydrocarbon receptor (AhR)-dependent transcriptional changes of selected target genes in human lung epithelial A549 cells incubated for 24 h, either with extractable organic matter (EOMs) from airborne particles <2.5 μm (PM2.5) collected at four localities from heavily polluted areas of the Czech Republic or two representative toxic polycyclic aromatic hydrocarbons (PAHs) present in EOMs, benzo[a]pyrene (B[a]P) and benzo[k]fluoranthene (B[k]F). Genotoxic effects were determined using DNA adduct analysis or analysis of expression of selected AhR-related genes involved in bioactivation of PAHs (CYP1A1, CYP1B1) and transcriptional repression (TIPARP). Our results suggested inhibition of formation of B[a]P-induced DNA adducts compared to individual B[a]P, probably attributable to competitive inhibition by other non-genotoxic EOM components. In contrast, induction of AhR target genes appeared not to be antagonized by the components of complex mixtures, as induction of CYP1A1, CYP1B1 and TIPARP transcripts reached maximum levels induced by PAHs.

The goal of our work is to integrate conventional mRNA expression profiles with miRNA expressions using the knowledge of their validated or predicted interactions in order to improve class prediction in genetically determined diseases. The raw mRNA and miRNA expression features become enriched or replaced by new aggregated features that model the mRNA-miRNA interaction. The proposed subtractive integration method is directly motivated by the inhibition/degradation models of gene expression regulation. The method aggregates mRNA and miRNA expressions by subtracting a proportion of miRNA expression values from their respective target mRNAs. The method is used to model the outcome or development of myelodysplastic syndrome, a blood cell production disease often progressing to leukemia. The reached results demonstrate that the integration improves classification performance when dealing with mRNA and miRNA profiles of comparable predictive power.

High-throughput genomic technologies have proved to be useful in the search for both genetic disease markers and more complex predictive and descriptive models. By the same token, it became obvious that accurate and interpretable models need to concern more than raw measurements taken at a single phase of gene expression. In order to reach a deeper understanding of the molecular nature of complexly orchestrated biological processes, all the available measurements and existing genomic knowledge need to be fused. In this paper, we introduce a tool for machine learning from heterogeneous gene expression data using prior knowledge. The tool is called miXGENE, it is elaborated upon in close connection with the biological departments that dispose of the above-mentioned data and have a strong interest in their integration within particular problem-oriented projects. The main idea is not merely to capture the transcriptional phase of gene expression quantified by the amount of messenger RNA~(mRNA). The increasing availability of microRNA~(miRNA) data asks for its concurrent analysis with the transcriptional data. Moreover, epigenetic data such as methylation measurements can help to explain unexpected transcriptional irregularities. miXGENE is an environment for building workflows that enable rapid prototyping of integrative molecular models.

Contemporary molecular biology deals with a wide and heterogeneous set of measurements to model and understand underlying biological processes including complex diseases. Machine learning provides a frequent approach to build such models. However, the models built solely from measured data often suffer from overfitting, as the sample size is typically much smaller than the number of measured features. In this paper, we propose a random forest-based classifier that minimizes this overfitting with the aid of prior knowledge in the form of a feature interaction network. We illustrate the proposed method in the task of disease classification based on measured mRNA and miRNA profiles complemented by the interaction network composed of the miRNA-mRNA target relations and mRNA-mRNA interactions corresponding to the interactions between their encoded proteins. We demonstrate that the proposed network-constrained forest employs prior knowledge to increase learning bias and consequently to improve classification accuracy, stability and comprehensibility of the resulting model. The experiments are carried out in the domain of myelodysplastic syndrome that we are concerned about in the long term. We validate our approach in the public domain of ovarian carcinoma, with the same data form. We believe that the idea of a network-constrained forest can straightforwardly be generalized towards arbitrary omics data with an available and non-trivial feature interaction network.

Onset and progression of a genetically conditioned disease depend not only on genes themselves, but mainly on their expression during transcriptional and proteosynthetic process. Monitoring gene expression merely at its transcription level often proves insufficient for an automated disease understanding and prediction. An integration of diverse high-throughput measurements and prior knowledge is needed to capture gene expression in a holistic way. In this paper, we apply a recent matrix factorization integration method to build a plausible and comprehensive predictive model of an outcome or progress of myelodysplastic syndrome, a blood production disease often progressing to leukemia. We propose an efficient learning methodology that enables to maximize predictive performance and keep the main assets of the original method. The resulting model shows a comparable predictive accuracy with a straightforward data integration method while being more understandable and compact. The identified gene expression regulatory units with the best predictive performance will be subject of further biological analysis.

Kapitola se zbývá metodami strojového učení pro využívání znalostí v procesu získávání znalostí.

Background: Analysis of gene expression data in terms of a priori defined gene sets has recently received significant attention as this approach typically yields more compact and interpretable results than those produced by traditional methods that rely on individual genes. The set level strategy can also be adopted with similar benefits in predictive classification tasks accomplished with machine learning algorithms. Initial studies into the predictive performance of set level classifiers have yielded rather controversial results. The goal of this study is to provide a more conclusive evaluation by testing various components of the set level framework within a large collection of machine learning experiments. Results: Genuine curated gene sets constitute better features for classification than sets assembled without biological relevance. For identifying the best gene sets for classification, the Global test outperforms the gene set methods GSEA and SAM GS as well as two generic feature selection methods. To aggregate expressions of genes into a feature value, the singular value decomposition (SVD) method as well as the SetSig technique improve on simple arithmetic averaging. Set level classifiers learned with 10 features constituted by the Global test slightly outperform baseline gene level classifiers learned with all original data features although they are slightly less accurate than gene level classifiers learned with a prior feature selection step. Conclusion: Set level classifiers do not boost predictive accuracy, however, they do achieve competitive accuracy if learned with the right combination of ingredients. Availability: Open-source, publicly available software was used for classifier learning and testing. The gene expression datasets and the gene set database used are also publicly available. The full tabulation of experimental results is available at http://ida.felk.cvut.cz/CESLT.

The availability of a great range of prior biological knowledge about the roles and functions of genes and gene-gene interactions allows us to simplify the analysis of gene expression data to make it more robust, compact and interpretable. Here, we objectively analyze the applicability of functional clustering for the identification of groups of functionally related genes. The analysis is performed in terms of gene expression classification and uses predictive accuracy as an unbiased performance measure. Features of biological samples that originally corresponded to genes are replaced by features that correspond to the centroids of the gene clusters and are then used for classifier learning. Using ten benchmark datasets, we demonstrate that functional clustering significantly outperforms random clustering without biological relevance. We also show that functional clustering performs comparably to gene expression clustering, which groups genes according to the similarity of their expression profiles. Finally, the suitability of functional clustering as a feature extraction technique is evaluated and discussed.

Background: Recently, we used cell-free assays to demonstrate the toxic effects of complex mixtures of organic extracts from urban air particles (PM2.5) collected in four localities of the Czech Republic (Ostrava-Bartovice, Ostrava-Poruba, Karvina and Trebon) which differed in the extent and sources of air pollution. To obtain further insight into the biological mechanisms of action of the extractable organic matter (EOM) from ambient air particles, human embryonic lung fibroblasts (HEL12469) were treated with the same four EOMs to assess changes in the genome-wide expression profiles compared to DMSO treated controls. Method: For this purpose, HEL cells were incubated with subtoxic EOM concentrations of 10, 30, and 60 μg EOM/ml for 24 hours and global gene expression changes were analyzed using human whole genome microarrays (Illumina). The expression of selected genes was verified by quantitative real-time PCR.

Bioinformatics is a field of study dealing with methods for storing, retrieving and analyzing gene and protein oriented biological data. High-throughput technologies like DNA sequencing or microarrays allow researchers to obtain large volumes of heterogeneous and mutually interacting data. Analysis and understanding of these data provides a natural application field for machine leasing algorithms. At the same time, bioinformatics is a scientific branch of such analytical complexity, data variety and abundance that it motivates further development of specialized learning algorithms such as co-clustering or multiple sequence alignment. This paper provides a brief overview of the topics and works discussed during my talk on machine learning applications in bioinformatics. The talk starts with a preview of fundamental bioinformatics analytical tasks solved by machine learning algorithms mentioning a few success stories. The second part summarizes the recent bioinformatics research carried out in my home research group, the Intelligent Data Analysis group of Czech Technical University.

BK polyomavirus infection is the important cause of virus-related nephropathy following kidney transplantation. BK virus reactivates in 30%-80% of kidney transplant recipients resulting in BK virus-related nephropathy in 1%-10% of cases. Currently, the molecular processes associated with asymptomatic infections in transplant patients infected with BK virus remain unclear. In this study we evaluate intrarenal molecular processes during different stages of BKV infection. The gene expression profiles of 90 target genes known to be associated with immune response were evaluated in kidney graft biopsy material using TaqMan low density array. Three patient groups were xamined: control patiens with no evidence of BK virus reactivation (n=11) , infected asymptomatic patiens (n=9) , and patiens with BK virus nephropathy (n = 10).

Analysis of gene expression data in terms of a priori-defined gene sets typically yields more compact and interpretable results than those produced by traditional methods that rely on individual genes. The set-level strategy can also be adopted in predictive classification tasks accomplished with machine learning algorithms. Here, sample features originally corresponding to genes are replaced by a much smaller number of features, each corresponding to a gene set and aggregating expressions of its members into a single real value. Classifiers learned from such transformed features promise better interpretability in that they derive class predictions from overall expressions of selected gene sets (e.g. corresponding to pathways) rather than expressions of specific genes. In a large collection of experiments we test how accurate such classifiers are compared to traditional classifiers based on genes.

The presented paper describes a method of text preprocessing improving the performance of sequential data mining applied in the task of gene interaction extraction from biomedical texts. The need of text preprocessing rises primarily from the fact, that the language encoded by any general word sequence is mostly not sequential. The method involves a number of heuristic language transformations, all together converting sentences into forms with higher degree of sequentiality. The core idea of enhancing sentence sequentiality results from the observation that the components constituting the semantical and grammatical content of sentences are not equally relevant for extracting a highly specific type of information. Experiments employing a simple sequential algorithm confirmed the usability of the proposed text preprocessing in the gene interaction extraction task.

The presented paper describes a method of text preprocessing improving the performance of sequential data mining applied in the task of gene interaction extraction from biomedical texts. The need of text preprocessing rises primarily from the fact, that the language encoded by any general word sequence is mostly not sequential. The method involves a number of heuristic language transformations, all together converting sentences into forms with higher degree of sequentiality. The core idea of enhancing sentence sequentiality results from the observation that the components constituting the semantical and grammatical content of sentences are not equally relevant for extracting a highly specific type of information. Experiments employing a simple sequential algorithm confirmed the usability of the proposed text preprocessing in the gene interaction extraction task.

V nedávno publikované studii jsme hodnotili toxické účinky komplexních směsí organických extraktů z prachových částic (PM2.5) ve vnějším ovzduší, odebraných ve 4 lokalitách České republiky s odlišnými zdroji znečištění ovzduší. V této části studie byly použity lidské embryonální plicní fibroblasty (HEL 12469) s cílem proniknout hlouběji do biologických mechanismů působení těchto směsí organických látek. Buňky HEL byly inkubovány 24 hodin se subtoxickými koncentracemi extraktů z PM2.5 ze všech studovaných lokalit. S použitím čipové technologie Illumina byla analyzována genová exprese na úrovni celého genomu, která byla pak porovnána s kontrolní úrovní genové exprese v buňkách inkubovaných jen s rozpouštědlem (DMSO). Exprese vybraných genů byla verifikována RT-PCR. Celkový počet významně deregulovaných genů byl úměrný dávce extraktu.

This paper presents an adaBoost approach for schizophrenia relapse prediction. The data for the adaBoost are extracted from patients answers to Early Warning Signs questionnaires sent regularly via mobile phone messages. The performance of the adaBoost algorithm is confronted with current ITAREPS system with sensitivity 0.65 and specificity 0.73. AdaBoost has the same sensitivity 0.65 but higher specificity 0.84 and is then ready to became the part of the ITAREPS care program.

We demonstrate how some recently developed techniques of set level gene expression data analysis may be exploited in the context of predictive classification of gene expression samples for the tasks of attribute selection and extraction. With four benchmark gene expression datasets, we empirically test the influence of these method on the predictive accuracy of constructed classification models in a comparative setting. Our results mainly indicate that gene set selection methods (SAM GS and the global test) can boost the predictive accuracy if used with caution.

Biomedical Named Entity Recognition (NER) is still a challenging problem. In this paper, we show that pattern mining techniques such as sequential pattern mining and sequential rules mining, can be useful to tackle this problem but present some limitations. That it is why we define a new kind of pattern called LSR patterns that offer an excellent trade-off between the high precision of sequential rules and the high recall of sequential patterns. We formalize the LSR pattern mining problem. We then show how LSR patterns enable us to successfully tackle biomedical NER problem. We report experiments carried out on real data sets that underline the relevance of our proposition.

This paper presents the web tool XGENE.ORG which facilitates the integration of gene expression measurements with background genomic information, in particular the gene ontology and KEGG pathways. The novelty of the proposed data fusion is in the introduction of working units at different levels of generality acting as sample features, replacing the commonly used gene units, consequently al-lowing for cross-genome (multi-platform) expression data analysis. The integration of different microarray platforms contributes to the robustness of knowledge extracted when single-platform samples are rare and facilitates inference of biological knowledge not constrained to single organisms.

Discovery of biologically interpretable knowledge from gene expression data is a crucial issue. Current gene data analysis is often based on global approaches such as clustering. An alternative way is to utilize local pattern mining techniques for global modelling and knowledge discovery. Nevertheless, moving from local patterns to models and knowledge is still a challenge due to the overwhelming number of local patterns and their summarization remains an open issue. This paper is an attempt to fulfill this need: thanks to recent progress in constraint-based paradigm, it proposes three data mining methods to deal with the use of local patterns by highlighting the most promising ones or summarizing them. Ideas at the core of these processes are removing redundancy, integrating background knowledge and recursive mining.

This chapter points out the role of genomic background knowledge in gene expression data mining. The authors demonstrate its application in several tasks such as relational descriptive analysis, constraint-based knowledge discovery, feature selection and construction or quantitative association rule mining. The chapter also accentuates diversity of background knowledge. In genomics, it can be stored in formats such as free texts, ontologies, pathways, links among biological entities and many others. The authors hope that understanding of automated integration of heterogeneous data sources helps researchers to reach compact and transparent as well as biologically valid and plausible results of their gene-expression data analysis.

We demonstrate a set-level approach to the integration of multiple platform gene expression data for predictive classification

Dimension reduction is the process of reducing the number of variables under consideration. In genomic classification it is widely applied because the high dimensionality of gene-expression data proved to decrease accuracy and comprehensibility of genomic classifiers. Simultaneously, contemporary genomics offers an opportunity to reach beyond the routine application of purely statistical dimension reduction techniques. Availability of a great variability of knowledge on gene roles, functions and gene-gene interactions allows to benefit from knowledge-based approaches to dimension reduction. This paper introduces and tests a feature-extraction algorithm that employs keywords affinity to define the gene similarity measure. This measure is used to form gene clusters whose medoids serve as new features. The features are of a reasonable number with statistically proven noise robustness and with anticipation of easy interpretability.

Current analyses of co-expressed genes are often based on global approaches such as clustering or bi-clustering. An alternative way is to employ local methods and search for patterns - sets of genes displaying specific expression properties in a set of situations. The main bottleneck of this type of analysis is twofold - computational costs and an overwhelming number of candidate patterns which can hardly be further exploited. A timely application of background knowledge available in literature databases, biological ontologies and other sources can help to focus on the most plausible patterns only. The paper proposes, implements and tests a flexible constraint-based framework that enables the effective mining and representation of meaningful over-expression patterns representing intrinsic associations among genes and biological situations.

Sekvenční data jsou důležitým zdrojem lékařských znalostí. Tato specifická data mohou vznikat řadou různých způsobů. V tomto článku na příkladu konkrétní studie prezentujeme obecné postupy pro jejich dolování. Jde o preventivní dlouhodobou studii atherosklerózy - data jsou výsledkem dvě dekády trvajícího sledování vývoje rizikových faktorů a přidružených jevů. Hlavním cílem je identifikovat časté sekvenční vzory, tj. opakující se časové jevy, a studovat jejich možnou souvislost s objevením jedné ze sledovaných kardiovaskulárních nemocí. Z širší škály dostupných metod se soustředíme na induktivní logické programování, které potenciální vzory vyjadřuje ve formě rysů v predikátové logice prvního řádu. Rysy jsou nejprve automaticky extrahovány a následně sdružovány do pravidel, která představují výstupní formu získané znalosti. Navržený postup je porovnán s tradičnějšími metodami publikovanými dříve. Jde o metodu posuvných oken a epizodní pravidla.

This paper deals with various method of gene expression data mining guided by genomic background knowledge, namely by the gene ontology.

Sequential data represent an important source of potentially new medical knowledge. However, this type of data is rarely provided in a format suitable for immediate application of conventional mining algorithms. This paper summarizes and compares three different sequential mining approaches, based respectively on windowing, episode rules and inductive logic programming. Windowing is one of the essential methods of data preprocessing, episode rules represent general sequential mining while inductive logic programming extracts first order features whose structure is determined by background knowledge. The three approaches are demonstrated and evaluated in terms of a case study STULONG. It is a longitudinal preventive study of atherosclerosis where the data consist of series of longterm observations recording the development of risk factors and associated conditions.

We compiled expression and pathway data related to a specific biological classification problem, into a relational database in Prolog, providing benchmarking material for ILP experimentation.

Continuous improvement of pharmaceutical manufacturing operations has not evolved at the same rate as it has in other industries. Although time series data are routinely collected as part of equipment control systems, the data are usually not thoroughly evaluated. This article investigates batch data, in process and release laboratory test data and time series data from granulation, fluid bed drying and coating operations in an effort to determine which parameters are most critical to the dissolution of a matrix release, solid oral dosage form of a poorly soluble drug.

Mining patterns under many kinds of constraints is a key point to successfully get new knowledge. In this paper, we propose an efficient new algorithm Music-dfs which soundly and completely mines patterns with various constraints from large data and takes into account external data represented by several heterogeneous datasets. Constraints are freely built of a large set of primitives and enable to link the information scattered in various knowledge sources. Efficiency is achieved thanks to a new closure operator providing an interval pruning strategy applied during the depth-first search of a pattern space. A genomic case study shows both the effectiveness of our approach and the added-value of background knowledge such as free texts or gene ontologies in discovery of meaningful patterns.

Regarding association rules, transcriptomic data represent a difficult mining context. First, the data are high-dimensional which asks for an algorithm scalable in the number of variables. Second, expression values are typically quantitative variables. This variable type further increases computational demands and may result in the output with a prohibitive number of redundant rules. Third, the data are often noisy which may also cause a large number of rules of little significance. In this paper we tackle the above-mentioned bottlenecks with an alternative approach to the quantitative association rule mining. The approach is based on simple arithmetic operations with variables and it outputs rules that do not syntactically differentiate from classical association rules. We also demonstrate the way in which apriori genomic knowledge can be used to prune the search space and reduce the amount of derived rules.

The general purpose of educational platforms is to provide students with information as well as with practical opportunities in order to help learners to acquire certain skills and to increase their active knowledge about a studied topic. However, different learners may have different characteristics, prior knowledge, motivation or needs. This diversity commonly requires the presentation of different information to different learners in a different format. That is why it is very important to develop adaptive educational systems which consider various aspects of individual students and tailor the learning process to meet the actual learner's needs. In this paper there are described basic demands put on an adaptive e-learning system. Basic types of barriers discouraging e-learning systems from being more used are mentioned. Comparison of several available e-learning systems (mostly open source ones) is done from the point of their adaptivity.

Automated extraction of links among biological entities from free biological texts has proven to be a difficult task. In this paper we propose and solve a modified task in which we extract the links from short textual gene summaries collected automatically from NCBI website. The main simplification lies in the fact that each summary is unambiguously attached to a single gene. The agent part of binary biological interactions is thus known by default, the goal is to identify meaningful target parts from the summary. The outcome is a structured representation of each summary that can be used as background knowledge in consequent mining of gene expression data. As the gene summaries highly interact with the other structural information resources provided by NCBI website, these resources can be used as an annotation tool and/or a feedback for performance optimization of the system being developed.

This paper presents a study of the task of classification and analysis of fanatic texts. The analyzed set of texts stems from an Arabic environment in Kuwait, where teachers and students were asked questions regarding various terrorist tendencies. The responses were assigned by a domain expert into one of three classes with respect to degree of fanaticism of their content. The main task was to grasp the implicit expert's knowledge and distinguish the documents according to their content. The paper deals with the bag-of-words representation of the documents. It applies learning algorithms that proved to work well in the field of text classification (TFIDF classifier, multinomial probabilistic model) as well as the random forest classifier that is well-known to cope with domains described by a large number of features. The associated task is to discover any knowledge helping to understand the domain.

This paper presents a research into the task of classification and analysis of fanatic texts. The analyzed set of texts stems from an Arabic environment in Kuwait, where teachers and students were asked regarding various terrorist tendencies. The responses were assigned by a domain expert into one of three classes with respect to degree of fanaticism of their content. The main task was to grasp the implicit expert's knowledge and distinguish the documents according to their content. The paper deals with the bag-of-words representation of the documents. It applies learning algorithms that proved to work well in the field of text classification (TFIDF classifier, multinomial probabilistic model) as well as the random forest classifier that is well-known to cope with domains described by a large number of features. The associated task is to discover any knowledge helping to understand the domain.

Mining patterns under many kinds of constraints is a key point to successfully get new knowledge. In this paper, we propose an efficient new algorithm Music-dfs which soundly and completely mines patterns with various constraints from large data and takes into account external data represented by several heterogeneous datasets. Constraints are freely built of a large set of primitives and enable to link the information scattered in various knowledge sources. Efficiency is achieved thanks to a new closure operator providing an interval pruning strategy applied during the depth-first search of a pattern space. A genomic case study shows both the effectiveness of our approach and the added-value of background knowledge such as free texts or gene ontologies in discovery of meaningful patterns.

The discovery of biologically interpretable knowledge from gene expression data is one of the largest contemporary genomic challenges. As large volumes of expression data are being generated, there is a great need for automated tools that provide the means to analyze them. However, the same tools can provide an overwhelming number of candidate hypotheses which can hardly be manually exploited by an expert. An additional knowledge helping to focus automatically on the most plausible candidates only can up-value the experiment significantly. Background knowledge available in literature databases, biological ontologies and other sources can be used for this purpose. In this paper we propose and verify a methodology that enables to effectively mine and represent meaningful over-expression patterns. Each pattern represents a bi-set of a gene group over-expressed in a set of biological situations.

Applications of Genetic Algorithms

Asociační pravidla byla prvoplánově navržena jako nástroj pro vyhledávání vazeb mezi binárními atributy. Přestože je přechod na domény obsahující i jiné typy atributů relativně přímočarý, může při něm docházet ke ztrátě užitečné informace. To platí zejména v případě atributů, jejichž hodnoty lze uspořádat - ordinálních atributů. Rozličné způsoby jejich transformace na binární atributy mohou vést ke kombinatorické explozi a v konečném důsledku i k velkému množství nevýznamných pravidel. Článek diskutuje alternativní přístup, ve kterém cedenty nejsou tvořeny konjunkcí literálů, ale jednoduchými operacemi zachovávajícími uspořádání.

Sequential data represent an important source of automatically mined and potentially new medical knowledge. They can originate in various ways. Within the presented domain they come from a longitudinal preventive study of atherosclerosis - the data consist of series of long-term observations recording the development of risk factors and associated conditions. The intention is to identify frequent sequential patterns having any relation to an onset of any of the observed cardiovascular diseases. This paper focuses on application of inductive logic programming. The prospective patterns are based on first-order features automatically extracted from the sequential data. The features are further grouped in order to reach final complex patterns expressed as rules. The presented approach is also compared with the approaches published earlier (windowing, episode rules).

This paper discusses issues of automated model tuning in order to obtain a proper definition of mutual case similarity in a specific medical domain. The main focus is on a reasonably time-consuming optimization of the parameters that determine case retrieval and further utilization in decision making/prediction.

Prediction is one of the tasks being solved by humans everyday. When encountering a complex and possibly dynamic domain containing a large number of possible dependencies, the predictive task can become too difficult to be solved by usual common sense reasoning. This article introduces Open Prediction System (OPS) - a system that helps to develop predictive models automatically. It represents a general predictive system applicable in a wide range of problem domains. The special attention is paid to the tasks of prediction in multivariate time series motivated by problems common for utility companies that distribute and control the transport of their applicable commodity. General issues of forecasting are discussed together with OPS predictive methodology implemented. Examples of case studies of such system are also included.

Článek popisuje postup při zpracování lékařských dat získaných dlouhodobým sledováním pacientů.

Při zpracování lékařských dat získaných dlouhodobým sledováním pacientů se často setkáváme s tím, že se rozsah dat pro jednotlivé pacienty výrazně liší. Data o sledovaném souboru jsou pak nejednotná, přičemž zařazení pacienta do cílové třídy může přímo souviset s počtem dostupných měření o tomto pacientovi. Ovšem u této hodnoty je nebezpečí, že jde o anachronický atribut a nelze se o ni opírat při konstrukci libovolného modelu sledovaných dat. Příspěvek navrhuje možný postup, jak předzpracovat výchozí soubor dat tak, aby byl tento zásadní problém odstraněn. Navržená metoda je ilustrována na případu dat ze studie STULONG

The paper is concerned with mining data which do not have uniform structure, e.g. they represent repeated measurements of different objects in the case the observation period is not fixed. The available data have to be transformed and preprocessed in such a way that uniform type of information is obtained about all the considered objects. This can be achieved e.g., by aggregation. But this process can bring in anachronistic variables, i.e., variables containing information which is not actually available in the data when a prediction is needed. The paper suggests a method how to preprocess considered type of data without falling into the trap of introducing anachronistic attributes.

The paper concerns mining data lacking the uniform structure. The data are collected from a number of objects during repeated measurements, all of which are tagged by a corresponding time. No attribute-valued machine learning algorithm can be applied directly on such data since the number of measurements is not fixed but it varies. The available data have to be transformed and preprocessed in such a way that a uniform type of information is obtained about all the considered objects. This can be achieved, e.g., by aggregation. But this process can introduce anachronistic variables, i.e., variables containing information which cannot be available at the moment when a prediction is needed. The paper suggests and tests a method how to preprocess the considered type of data without falling into a trap of introducing anachronistic attributes. The method is illustrated on a case study based on STULONG data.

This paper presents an application of ant colony optimisation and genetic algorithm to rescue operation planning. It considers the task as the multiple travelling salesmen problem and proposes suitable heuristics in order to improve the performance of the selected techniques. Then it applies the implemented solutions to a real data. The paper concludes with comparison of the implementations and discussion on the aspects of the utilisationof the proposed heuristics

Článek prezentuje výsledky pilotního projektu 2letého sledování pacientů po kardiochirurgických operacích včetně subjektivní složky zdraví. Použity jsou 2 dotazníky: Follow-up-(FU)-dotazník, zaměřený převážně na kardiologickou problematiku a dotazník kvality života Short Form 36 (SF-36). Výsledky jsou prezentovány na souboru 416 pacientů, kteří byli sledováni 3 měsíce po operaci, a dále pak 1 a 2 roky po operaci. Závěry mohou posloužit jako vodítko pro prevenci i pro péči o pacienty po kardiochirurgických výkonech. Mohou se rovněž stát podkladem pro rozsáhlejší studie podobného zaměření

Distribuční sítě jsou komplexními a dynamickými systémy. Pro jejich efektivní řízení je třeba důkladně porozumět základním mechanismům jejich chování. Správné zodpovězení otázek typu kolik? kdy? kam? přináší distributorům řadu finančních výhod. Obecně existuje řada způsobů jak stále nenahraditelné porozumění lidské doplnit o porozumění strojové. Článek prezentuje řešení založené na podpoře rozhodování realizované jako součást informačního systému distribuční společnosti. Bloky pro podporu rozhodování a řízení jsou tvořeny prediktivními a deskriptivními modely automaticky generovanými moderními metodami dolování dat.

This paper addresses the problem of on-line diagnosis of cavitation in centrifugal pumps. The paper introduces an application of the Open Prediction System (OPS) to cavitation diagnosis. The application of OPS results in an algorithmic framework for diagnosis of cavitation in centrifugal pumps. The diagnosis is based on repeated evaluation of a data scan providing full record of input signals which are observed for a fixed short period of time. Experimental verification of the algorithmic framework and the proposed methodology proved that a condition monitoring system built upon them is capable of diagnosing a wide range of cavitation conditions that can occur in a centrifugal pump, including the very early incipient cavitation.

The paper is focused on the analysis and design of multivariate time series prediction systems. It addresses mainly practical issues, the main contribution is the developed and implemented conceptual predictive methodology. It is based on designed data management structures that define basic data flow. Despite the fact that the methodology is inspired by problems common for utility companies that distribute and control the transport of their applicable commodity, it may be considered as a general methodology. The methodology is implemented in the form of a software tool. It is verified on a real-life prediction task - prediction of the daily gas consumption of regional gas utility companies.

Otevřený predikční systém (OPS) řeší úlohy predikce a dolování dat. Jádro systému nabízí několik algoritmů založených jak na principech strojového učení, tak i matematické statistiky. Mezi algoritmy najdeme analýzu singulárních čísel, neuronové sítě, metodu podpůrných vektorů a rozhodovací stromy. Celý systém je podpořen datovými strukturami a toky, které umožňují rychlou a snadnou definici problému, individuální přístup k jednotlivým algoritmům a přehledné vyhodnocení navržených modelů.

Aplication of three soft computing technigues to solve multiple travelling salesmen problem, self organized neural networks, genetic algorithm and ant colony optimization..

The ECML/PKDD data mining challenge concerns a dataset describing the data collected during a longitudinal study of atherosclerosis prevention on around 1400 middle-aged men (Stulong study). The data challenge entry from the Czech Technical University in Prague focuses on trend analysis in this dataset. Firstly, it proposes and verifies a preprocessing method based on windowing. The suggested approach guarantees that the identified trend aggregates are generated without falling into the trap of introducing anachronistic attributes. Secondly, it applies the windowing method to the Stulong dataset. Finally, it studies influence of these trend aggregates on a possible future development of cardiovascular diseases (CVDs).

Kapitola se zabývá otázkami kolaborativního dolování dat. Možné problémy jsou dokumentovány na příkladu konkrétního projektu plánování a alokace zdrojů v lázních. V závěru jsou navrženy znalostní struktury pro efektivní reprezentaci a sdílení informací v klíčových bodech procesu dolování dat

The project tries to integrate two modern fields: data mining and decision support. The project aim is to propose and standardize formats for data exchange and sharing between DM and DS systems as well as to suggest and test tools supporting integration. The paper focuses on the two case studies giving practical examples of such an integration.

Nowadays many real-world medical problems are being assessed with tools for automatic intelligent data analysis. Many different methods have been developed to improve the quality of analysis for specific domain. Application of method in specific domain requires special characteristics. For instance methods based on artificial neural network are capable of generalisation of nonlinearly separable problems but have poor explanatory power. Therefore we focused on methods which are capable of extracting knowledge in a form closer to human perception, e.g. methods that induce decision trees, classification rules, etc.

This paper presents a case study concerning scheduling and resource allocation issues at a spa. The paper is data-mining oriented. It discusses and describes how the history data can be used as a source for data-mining leading to discovery of rules or algorithms useful for prediction of resources requirements. In particular, we focused to identify groups of patients which appear frequently in the training set and which exhibit characteristic behavior or requirements of spa utilities. Then we predicted a set of health procedures to be passed for each member of such group. This approach resulted in a health procedure prediction algorithm satisfactory for early and convenient scheduling.

Snaha objevovat řád ve světě, který člověka obklopuje, je jedním z charakteristických projevů lidského ducha. Jak lze řád nalézt a popsat? Tuto otázku si kladou experimentální vědy vždy znovu a znovu, kdykoliv se snaží popsat nový fenomén. Před tímto problémem stál například Johann Kepler začátkem 17. století, když se jako dvorní matematik snažil odhalit řád v pozorováních, které nashromáždil Tycho Brahe při měření poloh planet v průběhu řady let.

Kapitola rekapituluje vybrané techniky strojového učení používané v úlohách prediktivního a deskriptivního dolování dat. Dále se zabývá otázkami předzpracování dat, hodnocení vytvářených modelů, popřípadě specifickými úlohami dolování dat (časová data, texty, multirelační data).

The 2003 ECML/PKDD data mining challenge concerns a dataset describing the data collected during a longitudinal study of atherosclerosis prevention on around 1400 middle-aged men at a number of Czech hospitals. The data challenge entry from the Czech Technical University in Prague takes an approach which is heavy on data preparation through well-defined data transformations. This document describes the special requirements of this data mining tasks, the transformations designed to meet them and it points to some interesting observations found in the studied dataset.

Článek shrnuje zkušenosti a poznatky získané v rámci kolaborativního projektu dobývání znalostí z dat. Zaměřuje se zejména na zefektivnění sdílení znalostí mezi spolupracujícími týmy.

Příspěvek popisuje využití metod dolování z dat (Data Mining) pro alokaci zdrojů, konkrétně pro predikci počtu léčebných procedur v lázních. Příspěvěk se zabývá zkoumáním dat, jejich předzpracováním a metodami pro predikci. Bylo otestováno několik různých metod od lineární regrese až po pokročilejší metody jako je třeba případové usuzování (Instance-Based Reasoning). Dosažené výsledky jsou porovnány a vyhodnoceny.

Strojové učení (ML - Machine Learning) dosahuje úspěchů v celé řadě praktických aplikací, což z něj činí populární a často využívanou třídu metod. Příspěvek se zabývá nejprve teoretickým pohledem na strojové učení a jeho algoritmy a metody. Obecně zhodnocuje, na jaké úrovni je současná adaptivní schopnost učících se systémů samotných a nakolik je vývoj nové aplikace podporován kooperativním úsilím znalostního inženýra a experta v dané problémové oblasti. V druhé části poté demonstruje některé úspěšné aplikace strojového učení realizované v rámci Gerstnerovy laboratoře. Hlavní pozornost je věnována úloze predikce spotřeby plynu, krátce je diskutována také úloha diagnostiky chybových stavů čerpadla.

This paper focuses on development and optimized tuning of a model predicting risks related to heart interventions of several types. The model is based on representative data sets collected in the Merged National Registry (MNR) on Cardiovascular Interventions. The central attention is paid to an instance-based reasoning model and its tuning. In particular, the paper presents and discusses benefits of utilization of a genetic algorithm with limited convergence for this purpose.

Využití metod strojového učení v multiagentních systémech

Multi-agent systems are usually very complex in their structure and functionality. fn most of the application tasks, it is,difficult or sometimes impossible to determine exactly and correctly behavior and activities of a multi-agent system during its design. Therefore it is important to find a way how to improve system's activity during its operation. This can be achieved by learning agents which modify their behaviour according to their experience. There have to be studied and developed new methods of machine learning which will prove useful for this purpose. The paper reviews the basic problems of learning in multi-agent systems and some approaches applied for their solution.

This paper deals with utilization of background knowledge in machine learning methods.