Drug-induced gene expression profiles are rich information sources that can help to measure the effect of a drug on the transcriptional state of cells. However, the available experimental data only covers a limited set of conditions such as treatment time, dosages, and cell lines. This poses a challenge for neural network models to learn embeddings that can be generalized to new experimental conditions. In this project, we focus on the cell line as the confounder variable and train an Adversarial Neural Network to extract transcriptional effects that are conserved across multiple cell lines, and can thus be more confidently generalized to the biological setting of interest. Additionally, we investigate several methods to test whether our approach can simultaneously learn biologically valid embeddings and deconfound the effect of cell lines on the data distribution

Since the initial genome sequencing projects along with the recent advances on technology, molecular biology and large scale transcriptome analysis result in data accumulation at a large scale. These data have been provided in different platforms and come from different laboratories therefore, there is a need for compilation and comprehensive analysis. In this thesis, we addressed the automatization of biological data acquisition and integration from these non-uniform data using machine learning techniques. We focused on two different mining studies in the scope of this thesis. In the first study, we worked on characterizing expression patterns of housekeeping genes. We described methodologies to compare measures of housekeeping genes with non-housekeeping genes. In the second study, we proposed a novel framework, bi-k-bi clustering, for finding association rules of gene pairs that can easily operate on large scale and multiple heterogeneous data sets. Results in both studies showed consistency and relatedness with the available literature. Furthermore, our results provided some novel insights waiting to be experimented by the biologists.

As genes with similar expression pattern are very likely having the same biological function, cluster analysis becomes an important tool to understand and predict gene functions from gene expression profi les. In many situations, each gene expression profi le only contains a few data points. Directly applying traditional clustering algorithms to such short gene expression profi les does not yield satisfactory results. Developing clustering algorithms for short gene expression profi les is necessary. In this thesis, two novel methods are developed for clustering short gene expression pro files. The fi rst method, called the network-based clustering method, deals with the defect of short gene expression profi les by generating a gene co-expression network using conditional mutual information (CMI), which measures the non-linear relationship between two genes, as well as considering indirect gene relationships in the presence of other genes. The network-based clustering method consists of two steps. A gene co-expression network is firstly constructed from short gene expression profi les using a path consistency algorithm (PCA) based on the CMI between genes. Then, a gene functional module is identi ed in terms of cluster cohesiveness. The network-based clustering method is evaluated on 10 large scale Arabidopsis thaliana short time-course gene expression profi le datasets in terms of gene ontology (GO) enrichment analysis, and compared with an existing method called Clustering with Over-lapping Neighbourhood Expansion (ClusterONE). Gene functional modules identi ed by the network-based clustering method for 10 datasets returns target GO p-values as low as 10-24, whereas the original ClusterONE yields insigni cant results. In order to more speci cally cluster gene expression profi les, a second clustering method, namely the protein-protein interaction (PPI) integrated clustering method, is developed. It is designed for clustering short gene expression profi les by integrating gene expression profi le patterns and curated PPI data. The method consists of the three following steps: (1) generate a number of prede ned profi le patterns according to the number of data points in the profi les and assign each gene to the prede fined profi le to which its expression profi le is the most similar; (2) integrate curated PPI data to refi ne the initial clustering result from (1); (3) combine the similar clusters from (2) to gradually reduce cluster numbers by a hierarchical clustering method. The PPI-integrated clustering method is evaluated on 10 large scale A. thaliana datasets using GO enrichment analysis, and by comparison with an existing method called Short Time-series Expression Miner (STEM). Target gene functional clusters identi ed by the PPI-integrated clustering method for 10 datasets returns GO p-values as low as 10-62, whereas STEM returns GO p-values as low as 10-38. In addition to the method development, obtained clusters by two proposed methods are further analyzed to identify cross-talk genes under fi ve stress conditions in root and shoot tissues. A list of potential abiotic stress tolerant genes are found.

O Diabetes Mellitus tipo 1 (DM1) resulta de um ataque autoimune contra as células pancreáticas, extinguindo a produção de insulina e levando à hiperglicemia. Evidências indicam uma associação entre o estresse oxidativo (que pode causar danos no DNA) e o DM1, sendo que apenas alguns trabalhos da literatura relataram a expressão de genes relacionados à respostas ao estresse oxidativo e reparo do DNA em DM1. Ainda, os microRNAs (reguladores pós-transcricionais da expressão gênica) estão envolvidos em vários processos biológicos e condições patológicas, mas informação sobre a expressão dos microRNAs em DM1 ainda é escassa. A fim de proporcionar um melhor entendimento sobre as vias de regulação de genes participantes de processos biológicos relevantes para o DM1, o presente estudo consistiu em analisar os perfis de expressão gênica (método de microarranjos) de microRNAs e de mRNAs (bem como de algumas proteínas) provenientes de células mononucleares do sangue periférico (PBMCs, do inglês peripheral blood mononuclear cells) de pacientes DM1 (n=19) em comparação com indivíduos sadios não diabéticos (n=11), dando maior enfoque a genes associados à resposta ao estresse oxidativo e reparo do DNA. Os resultados de expressão obtidos pelo método de microarranjos apontaram 44 microRNAs diferencialmente expressos (35 induzidos e nove reprimidos) nos pacientes DM1 e esses microRNAs apresentaram grande especificidade ao estratificar pacientes DM1 dos controles, incluindo hsa-miR-101, hsa-miR148a, hsa-miR-27b e hsa-miR-424, cujos dados de expressão foram confirmados por qRT-PCR. A análise funcional dos genes-alvo dos microRNAs, tanto dos induzidos quanto dos reprimidos, apontou 22 e 12 vias KEGG significativamente enriquecidas, respectivamente, incluindo vias relacionadas ao câncer. Com relação à análise de expressão de mRNAS, 277 genes diferencialmente expressos foram identificados nos pacientes DM1, sendo que 52% deles são potenciais alvos dos microRNAs diferencialmente expressos nos pacientes DM1. Dentre esses alvos foram encontrados genes candidatos ao desenvolvimento da doença, assim como genes implicados nos processos biológicos resposta ao estresse oxidativo e reparo do DNA, como UCP3, PTGS2, ATF3, FOSB, DUSP1 e TNFAIP3, cujos dados de expressão foram confirmados por qRT-PCR. Já a análise de grupos gênicos identificou 49 e 55 grupos gênicos significativamente expressos e enriquecidos em pacientes DM1, respectivamente, destacando-se vias relacionadas à sinalização apoptótica, resposta ao hidroperóxido, reparo do DNA por recombinação homóloga e resposta ao estresse do retículo endoplasmático. Quanto aos dados de expressão proteica (western blotting), PTGS2 e ATF3 não apresentaram níveis de expressão detectáveis em nenhum dos dois grupos estudados, enquanto que para DUSP1 não foi observada diferença estatisticamente significativa entre os grupos, apesar de os três genes se apresentarem induzidos em pacientes DM1. Os resultados do ensaio do gene repórter da luciferase demonstraram a ocorrência da interação entre hsa-miR-148a e DUSP1 em meio celular. Essa evidência aliada aos dados de western blotting, sugerem a possibilidade de hsa-miR-148a atuar na repressão traducional de DUSP1. Em conjunto, os resultados do presente estudo indicaram perfis distintos de expressão de microRNAs e mRNAs em PBMCs de pacientes DM1 comparados a indivíduos sadios, sendo que adicionalmente, dados inéditos relacionados à interação microRNAs-mRNAs em DM1 foram obtidos, principalmente associados à resposta ao estresse oxidativo e reparo do DNA, sugerindo um distúrbio na rede microRNA-alvo em pacientes DM1. / Type 1 Diabetes Mellitus (T1DM) results from an autoimmune attack against the pancreatic cells, ceasing insulin production, which causes hyperglycemia. Although associations between oxidative stress, which can cause DNA damage, and T1DM have been demonstrated, only a few studies have reported differential expression of genes associated with response to oxidative stress and DNA repair in T1DM patients. Moreover, microRNAs (post-transcriptional regulators of gene expression) are implicated in many biological processes and pathological conditions; however, only scarce information is available in the literature concerning the expression of microRNAs in T1DM. In order to better understand the regulatory pathways involved in biological processes that are relevant to T1DM, we aimed to investigate the microRNA and mRNA transcriptional expression profiles by microarray analysis (as well as expression of selected proteins) in peripheral blood mononuclear cells (PBMCs) from T1DM patients (n=19) compared with healthy non-diabetic individuals (n=11), emphasizing genes related to response to oxidative stress and DNA repair. Microarray expression results indicated 44 differentially expressed microRNAs (35 up- and nine down-regulated) in T1DM patients, with those microRNAs possessing a discriminatory power to clearly stratify the patients from the controls, including hsa-miR-101, hsa-miR148a, hsa-miR-27b, and hsa-miR-424, whose expression data were confirmed by qRT-PCR. Functional annotation analysis performed on the predicted targets of the differentially expressed microRNAs pointed 22 and 12 annotated KEGG pathways for the overexpressed and repressed microRNAs, respectively, many of them related to cancer. Regarding mRNA microarray results, we detected 277 differentially expressed genes in T1DM patients, with 52% of them being potential targets of the differentially expressed microRNAs in T1DM patients. Among these targets, we identified candidate genes for T1DM as well as genes involved in the biological processes response to oxidative stress and DNA repair, such as UCP3, PTGS2, ATF3, FOSB, DUSP1 and TNFAIP3, whose expression data were confirmed by qRT-PCR. Furthermore, out of the 49 and 55 significantly expressed/enriched gene sets in T1DM patients, respectively, five pathways related to apoptotic signaling, response to hydroperoxide, DNA repair via homologous recombination, and response to endoplasmic reticulum stress were of interest for the present work. Concerning protein expression results (western blotting), PTGS2 and ATF3 expression was not detected for either the patient or the control group, while significant difference in DUSP1 expression was not observed between the two groups, although the corresponding mRNAs of those genes were found induced. Regarding the luciferase assay, our results demonstrated that the interaction between hsa-miR-148a and DUSP1 occurs in the cellular milieu. Therefore, these findings together with those western blotting results suggest that hsa-miR-148a could play a role in DUSP1 translational repression. Altogether, our results indicate distinctive microRNA and mRNA expression profiles in PBMCs from T1DM patients relative to healthy non-diabetic individuals. Furthermore, we have provided novel data regarding microRNA-mRNA interactions in T1DM, in particular involving genes associated with response to oxidative stress and DNA repair, suggesting a perturbation in the microRNA-target network in T1DM patients.

O Diabetes Mellitus tipo 1 (DM1) resulta de um ataque autoimune contra as células pancreáticas, extinguindo a produção de insulina e levando à hiperglicemia. Evidências indicam uma associação entre o estresse oxidativo (que pode causar danos no DNA) e o DM1, sendo que apenas alguns trabalhos da literatura relataram a expressão de genes relacionados à respostas ao estresse oxidativo e reparo do DNA em DM1. Ainda, os microRNAs (reguladores pós-transcricionais da expressão gênica) estão envolvidos em vários processos biológicos e condições patológicas, mas informação sobre a expressão dos microRNAs em DM1 ainda é escassa. A fim de proporcionar um melhor entendimento sobre as vias de regulação de genes participantes de processos biológicos relevantes para o DM1, o presente estudo consistiu em analisar os perfis de expressão gênica (método de microarranjos) de microRNAs e de mRNAs (bem como de algumas proteínas) provenientes de células mononucleares do sangue periférico (PBMCs, do inglês peripheral blood mononuclear cells) de pacientes DM1 (n=19) em comparação com indivíduos sadios não diabéticos (n=11), dando maior enfoque a genes associados à resposta ao estresse oxidativo e reparo do DNA. Os resultados de expressão obtidos pelo método de microarranjos apontaram 44 microRNAs diferencialmente expressos (35 induzidos e nove reprimidos) nos pacientes DM1 e esses microRNAs apresentaram grande especificidade ao estratificar pacientes DM1 dos controles, incluindo hsa-miR-101, hsa-miR148a, hsa-miR-27b e hsa-miR-424, cujos dados de expressão foram confirmados por qRT-PCR. A análise funcional dos genes-alvo dos microRNAs, tanto dos induzidos quanto dos reprimidos, apontou 22 e 12 vias KEGG significativamente enriquecidas, respectivamente, incluindo vias relacionadas ao câncer. Com relação à análise de expressão de mRNAS, 277 genes diferencialmente expressos foram identificados nos pacientes DM1, sendo que 52% deles são potenciais alvos dos microRNAs diferencialmente expressos nos pacientes DM1. Dentre esses alvos foram encontrados genes candidatos ao desenvolvimento da doença, assim como genes implicados nos processos biológicos resposta ao estresse oxidativo e reparo do DNA, como UCP3, PTGS2, ATF3, FOSB, DUSP1 e TNFAIP3, cujos dados de expressão foram confirmados por qRT-PCR. Já a análise de grupos gênicos identificou 49 e 55 grupos gênicos significativamente expressos e enriquecidos em pacientes DM1, respectivamente, destacando-se vias relacionadas à sinalização apoptótica, resposta ao hidroperóxido, reparo do DNA por recombinação homóloga e resposta ao estresse do retículo endoplasmático. Quanto aos dados de expressão proteica (western blotting), PTGS2 e ATF3 não apresentaram níveis de expressão detectáveis em nenhum dos dois grupos estudados, enquanto que para DUSP1 não foi observada diferença estatisticamente significativa entre os grupos, apesar de os três genes se apresentarem induzidos em pacientes DM1. Os resultados do ensaio do gene repórter da luciferase demonstraram a ocorrência da interação entre hsa-miR-148a e DUSP1 em meio celular. Essa evidência aliada aos dados de western blotting, sugerem a possibilidade de hsa-miR-148a atuar na repressão traducional de DUSP1. Em conjunto, os resultados do presente estudo indicaram perfis distintos de expressão de microRNAs e mRNAs em PBMCs de pacientes DM1 comparados a indivíduos sadios, sendo que adicionalmente, dados inéditos relacionados à interação microRNAs-mRNAs em DM1 foram obtidos, principalmente associados à resposta ao estresse oxidativo e reparo do DNA, sugerindo um distúrbio na rede microRNA-alvo em pacientes DM1. / Type 1 Diabetes Mellitus (T1DM) results from an autoimmune attack against the pancreatic cells, ceasing insulin production, which causes hyperglycemia. Although associations between oxidative stress, which can cause DNA damage, and T1DM have been demonstrated, only a few studies have reported differential expression of genes associated with response to oxidative stress and DNA repair in T1DM patients. Moreover, microRNAs (post-transcriptional regulators of gene expression) are implicated in many biological processes and pathological conditions; however, only scarce information is available in the literature concerning the expression of microRNAs in T1DM. In order to better understand the regulatory pathways involved in biological processes that are relevant to T1DM, we aimed to investigate the microRNA and mRNA transcriptional expression profiles by microarray analysis (as well as expression of selected proteins) in peripheral blood mononuclear cells (PBMCs) from T1DM patients (n=19) compared with healthy non-diabetic individuals (n=11), emphasizing genes related to response to oxidative stress and DNA repair. Microarray expression results indicated 44 differentially expressed microRNAs (35 up- and nine down-regulated) in T1DM patients, with those microRNAs possessing a discriminatory power to clearly stratify the patients from the controls, including hsa-miR-101, hsa-miR148a, hsa-miR-27b, and hsa-miR-424, whose expression data were confirmed by qRT-PCR. Functional annotation analysis performed on the predicted targets of the differentially expressed microRNAs pointed 22 and 12 annotated KEGG pathways for the overexpressed and repressed microRNAs, respectively, many of them related to cancer. Regarding mRNA microarray results, we detected 277 differentially expressed genes in T1DM patients, with 52% of them being potential targets of the differentially expressed microRNAs in T1DM patients. Among these targets, we identified candidate genes for T1DM as well as genes involved in the biological processes response to oxidative stress and DNA repair, such as UCP3, PTGS2, ATF3, FOSB, DUSP1 and TNFAIP3, whose expression data were confirmed by qRT-PCR. Furthermore, out of the 49 and 55 significantly expressed/enriched gene sets in T1DM patients, respectively, five pathways related to apoptotic signaling, response to hydroperoxide, DNA repair via homologous recombination, and response to endoplasmic reticulum stress were of interest for the present work. Concerning protein expression results (western blotting), PTGS2 and ATF3 expression was not detected for either the patient or the control group, while significant difference in DUSP1 expression was not observed between the two groups, although the corresponding mRNAs of those genes were found induced. Regarding the luciferase assay, our results demonstrated that the interaction between hsa-miR-148a and DUSP1 occurs in the cellular milieu. Therefore, these findings together with those western blotting results suggest that hsa-miR-148a could play a role in DUSP1 translational repression. Altogether, our results indicate distinctive microRNA and mRNA expression profiles in PBMCs from T1DM patients relative to healthy non-diabetic individuals. Furthermore, we have provided novel data regarding microRNA-mRNA interactions in T1DM, in particular involving genes associated with response to oxidative stress and DNA repair, suggesting a perturbation in the microRNA-target network in T1DM patients.

バイオマーカーはある特定の疾病や体質に関与する生体内分子のことであり，個別化医療の実現に貢献する．時系列の遺伝子発現プロファイルはバイオマーカーの探索に有用であるが，サンプルサイズに対して遺伝子数が多い，遺伝子間の多重共線性，時系列の複雑さなどの課題があった．本研究では，これらの課題に対応した時系列の遺伝子プロファイルを用いたバイオマーカーの探索手法を提案し，その有効性を示した． / Biomarkers contribute to designing the therapy. Time-course gene expression profiles were useful for biomarker discovery. However, biomarker discovery using time-course gene expression profiles suffered from three difficulties, including (1) high dimensional data with small sample size and a large number of genes, (2) multi-collinearity among genes, and (3) complexity of time-course. To solve these problems, expanding elastic net for consistent differentiation (eENCD) and consolidating probabilities of multiple time points (CPMTP) as new biomarker discovery methods using time-course gene expression profiles, were proposed and evaluated. / 博士(工学) / Doctor of Philosophy in Engineering / 同志社大学 / Doshisha University