Technological advances in next-generation sequencing and biomedical imaging have led to a rapid increase in biomedical data dimension and acquisition rate, which is challenging the conventional data analysis strategies. Modern machine learning techniques promise to leverage large data sets for finding hidden patterns within them, and for making accurate predictions. This dissertation aims to design novel machine learning-based models to transform biomedical big data into valuable biological insights. The research presented in this dissertation focuses on three bioinformatics domains: splice junction classification, gene regulatory network reconstruction, and lesion detection in mammograms. A critical step in defining gene structures and mRNA transcript variants is to accurately identify splice junctions. In the first work, we built the first deep learning-based splice junction classifier, DeepSplice. It outperforms the state-of-the-art classification tools in terms of both classification accuracy and computational efficiency. To uncover transcription factors governing metabolic reprogramming in non-small-cell lung cancer patients, we developed TFmeta, a machine learning approach to reconstruct relationships between transcription factors and their target genes in the second work. Our approach achieves the best performance on benchmark data sets. In the third work, we designed deep learning-based architectures to perform lesion detection in both 2D and 3D whole mammogram images.

Ultra-deep RNA sequencing has become a powerful approach for genome-wide analysis of pre-mRNA alternative splicing. We develop MATS (Multivariate Analysis of Transcript Splicing), a Bayesian statistical framework for flexible hypothesis testing of differential alternative splicing patterns on RNA-Seq data. MATS uses a multivariate uniform prior to model the between-sample correlation in exon splicing patterns, and a Markov chain Monte Carlo (MCMC) method coupled with a simulation-based adaptive sampling procedure to calculate the P value and false discovery rate (FDR) of differential alternative splicing. Importantly, the MATS approach is applicable to almost any type of null hypotheses of interest, providing the flexibility to identify differential alternative splicing events that match a given user-defined pattern. We evaluated the performance of MATS using simulated and real RNA-Seq data sets. In the RNA-Seq analysis of alternative splicing events regulated by the epithelial-specific splicing factor ESRP1, we obtained a high RT-PCR validation rate of 86% for differential alternative splicing events with a MATS FDR of < 10%. Additionally, over the full list of RT-PCR tested exons, the MATS FDR estimates matched well with the experimental validation rate. Our results demonstrate that MATS is an effective and flexible approach for detecting differential alternative splicing from RNA-Seq data.

Lung cancer is the second most common cancer type and is associated with high mortality rates. The survival rate for lung cancer patients has increased slowly in the last decade mainly as the result of the development of novel targeted and immune therapies. However, non-small cell lung cancer patients lacking known or actionable driver mutations and small cell lung cancer patients with recurrent disease are still in urgent need of new therapies. Drug repurposing is an efficient way to identify new therapies since it uses clinically relevant small molecule drugs. Determination of off-targets of small molecules is a novel approach towards drug repurposing as unintended targets can play important roles for a new clinical application. Here we apply functional proteomics and systems pharmacology approaches to determine target profiles and differential mechanisms of action, independent of the intended targets, of small molecules with similar chemical structures. Using chemical proteomics, we elucidated the differential target profiles of two clinical CDK4/6 inhibitors: palbociclib and ribociclib. We observed that palbociclib, but not ribociclib, is a dual protein and lipid kinase inhibitor that in addition to the intended cell cycle pathway modulates the PI3K/AKT and autophagy pathways. Furthermore, we investigated the off-targets of two MET/VEGFR inhibitors, foretinib and cabozantinib. Foretinib, but not cabozantinib, was found to reduce cell viability and induce cell cycle arrest and apoptosis in lung cancer cell lines. Using a systems pharmacology approach, which included chemical proteomics, phosphoproteomics and RNA-Seq, integrated data analysis and subsequent functional validation revealed a complex polypharmacology mechanism of action of foretinib, which involves the simultaneous inhibition of MEK, AURKB and FER protein kinases. Because AURKB is an important protein kinase for the proliferation of MYC-amplified small cell lung cancer cells, we were able to design a drug combination of foretinib with barasertib, a much more potent AURKB inhibitor, that enhanced specifically the cell death of MYC-amplified small cell lung cancer cells. In summary, we show that small structural changes of closely related clinical drugs can result in pronounced differences in their target profiles and anticancer activities through differential polypharmacology mechanisms and that an integrated systems pharmacology approach can identify new repurposing opportunities for these drugs in cancers with high unmet medical need.

La segmentation du rhombencéphale désigne le processus de subdivision rostro-caudale du cerveau postérieur embryonnaire en 7 compartiments cellulaires, appelés rhombomères (r), qui constituent des unités de développement et d'expression génique. Le facteur de transcription à doigts à zinc Krox20 joue un rôle central dans ce processus en couplant la formation et la spécification des rhombomères 3 et 5. Trois éléments régulateurs contrôlant l'expression de Krox20 ont été caractérisés. Les éléments B et C, actifs respectivement dans r5 et r3-r5, sont impliqués dans le démarrage de l'expression de Krox20 dont l'amplification et le maintien sont ensuite assurés par l'élément autorégulateur A. L'élément C constitue donc une clé d'investigation pour élucider les mécanismes responsables de l'expression régionalisée de Krox20 dans r3. Afin de poursuivre la caractérisation du contrôle transcriptionnel de Krox20 dans r3, j'ai entrepris une analyse fonctionnelle des séquences de l'élément C par mutagénèse et mis en évidence plusieurs blocs nécessaires à son activité. J'ai également établi le répertoire transcriptomique exhaustif, quantitatif et différentiel du rhombencéphale murin en début de segmentation par RNA-Seq et j'ai pu ainsi identifier plusieurs régulateurs de l'activité de l'élément C. Enfin, j'ai réalisé l'analyse du mutant murin d'excision de l'élément C et révélé de nouvelles propriétés du contrôle transcriptionnel de Krox20, dont la coopération en cis entre les éléments A et C. Ces travaux illustrent comment la combinaison de différentes approches permet d'élucider un mécanisme de régulation transcriptionnel complexe jouant un rôle essentiel dans la spécification cellulaire

There is limited information currently available regarding the underlying physiological responses and molecular mechanisms of osmoregulation, acetate metabolism [in relation to the synthesis of glycerolipids, polyunsaturated fatty acids (PUFA), and ichthyotoxins], and transport in Prymnesium parvum N. Carter, a microalga that causes devastating harmful algal blooms (HAB) worldwide. This dissertation examines gene expression under environmental conditions that are associated with HAB formation, including phosphate limitation and low salinity, using microarrays and RNA sequencing (RNA-Seq). A comparative fatty acid methyl ester (FAME) analysis at 30 vs. 5 practical salinity units (psu) was performed to gain additional insight into acetate metabolism. The RNA-Seq analysis included a de novo assembly of the P. parvum transcriptome, generating 47,289 transcripts, of which 35.4% were identifiable. This permitted the evaluation of the expression of many more genes compared with the microarray analysis, which examined ~3,500 genes. Relevant candidate genes identified included those whose products are involved in osmolyte production, salinity stress, and ion transport. With respect to the putative synthesis of polyketide ichthyotoxins, 32 different polyketide synthase (PKS) transcripts were identified in the transcriptome assembly, none of which were differentially expressed. Hemolysin and monogalactosyldiacylglycerol synthase were downregulated at 30 vs. 5 psu, suggesting the increased presence of additional ichthyotoxins at the lower salinity. Evidence for several PUFA synthesis pathways was also revealed. Fatty acid compositions were largely similar at the two salinities, containing relatively prominent quantities of the PUFA stearidonic acid, but compositions varied among strains. The transcription of genes whose products are associated with vesicular transport was elevated, and higher levels of extracellular prymnesins were observed in HAB-forming conditions. Thus, with regard to acetate metabolism, I have revealed evidence for the post-transcriptional regulation of the production of prymnesins and the contributory effects of hemolysin, monogalactosyldiacylglycerol, and PUFA towards ichthyotoxicity. Further, I propose that toxin transport is triggered in HAB-forming conditions, in which the toxins are actively being excreted. Collectively, these data shed light on the transcriptional responses that occur following alterations in phosphate availability and salinity, including those associated with the synthesis and delivery of a number of potential ichthyotoxins from P. parvum. / text

Transcriptional regulation is a complicated process controlled by numerous factors such as transcription factors (TFs), chromatin remodeling enzymes, nucleosomes, post-transcriptional machineries, and cis-acting DNA sequence. I explored the complex transcriptional regulation in eukaryotes through three distinct studies to comprehensively understand the functional genomics at various steps. Although a variety of high throughput approaches have been developed to understand this complex system on a genome wide scale with high resolution, a lack of accurate and comprehensive annotation transcription start sites (TSS) and polyadenylation sites (PAS) has hindered precise analyses even in Saccharomyces cerevisiae, one of the simplest eukaryotes. We developed Simultaneous Mapping Of RNA Ends by sequencing (SMORE-seq) and identified the strongest TSS and PAS of over 90% of yeast genes with single nucleotide resolution. Owing to the high accuracy of TSS identified by SMORE-seq, we detected possibly mis-annotated 150 genes that have a TSS downstream of the annotated start codon. Furthermore, SMORE-seq showed that 5’-capped non-coding RNAs were highly transcribed divergently from TATA-less promoters in wild-type cells under normal conditions. Mapping of DNA-protein interactions is essential to understanding the role of TFs in transcriptional regulation. ChIP-seq is the most widely used method for this purpose. However, careful attention has not been given to technical bias reflected in final target calling due to many experimental steps of ChIP-seq including fixation and shearing of chromatin, immunoprecipitation, sequencing library construction, and computational analysis. While analyzing large-scale ChIP-seq data, we observed that unrelated proteins appeared to bind to the gene bodies of highly transcribed genes across datasets. Control experiments including input, IgG ChIP in untagged cells, and the Golgi factor Mnn10 ChIP also showed the strong binding at the same loci, indicating that the signals were obviously derived from bias that is devoid of biological meaning. In addition, the appearance of nucleosomal periodicity in ChIP-seq data for proteins localizing to gene bodies is another bias that can be mistaken for false interactions with nucleosomes. We alleviated these biases by correcting data with proper negative controls, but the biases could not be completely removed. Therefore, caution is warranted in interpreting the results from ChIP-seq. Nucleosome positioning is another critical mechanism of transcriptional regulation. Global mapping of nucleosome occupancy in S. cerevisiae strains deleted for chromatin remodeling complexes has elucidated the role of these complexes on a genome wide scale. In this study, loss of chromodomain helicase DNA binding protein 1 (Chd1) resulted in severe disorganization of nucleosome positioning. Despite the difficulties of performing ChIP-seq for chromatin remodeling complexes due to their transient and dynamic localization on chromatin, we successfully mapped the genome-wide occupancy of Chd1 and quantitatively showed that Chd1 co-localizes with early transcription elongation factors, but not late transcription elongation factors. Interestingly, Chd1 occupancy was independent of the methylation levels at H3K36, indicating the necessity of a new working model describing Chd1 localization.

Characterization of the RNA transcriptome by next-generation sequencing can produce an unprecedented yield of information that provides novel biologic insights. I describe four approaches for sequencing different aspects of the transcriptome and provide computational tools to analyze the resulting data. Methods that query the dynamic range of gene expression, low expressing transcripts, micro RNA levels, and start-site usage of transcripts are described.

Somatic cells can be reprogrammed to induced pluripotent stem cells (iPSCs) by the ectopic expression of defined factors. iPSCs hold great promise for pharmaceutical screening and regenerative medicine but the mechanism of reprogramming is not well understood. This work examines a component process of reprogramming that is the mesenchymal-to-epithelial transition (MET), an important step in the generation of iPS cells. In this thesis I demonstrate a connection between BMP signaling and the reprogramming factor Klf4 in the activation of the MET expression program. Using ChIP-Seq I mapped the binding of Klf4 and BMP Smads across the genome and linked their co-binding to a MET expression program determined by RNA-Seq. My work uncovers a thus-far unreported interaction between Klf4 and BMP signaling in cellular epithelialization that can directly improve the technical methods of reprogramming and have important implications for the induction of epithelial tissues in general.

Somatic cells can be reprogrammed to induced pluripotent stem cells (iPSCs) by the ectopic expression of defined factors. iPSCs hold great promise for pharmaceutical screening and regenerative medicine but the mechanism of reprogramming is not well understood. This work examines a component process of reprogramming that is the mesenchymal-to-epithelial transition (MET), an important step in the generation of iPS cells. In this thesis I demonstrate a connection between BMP signaling and the reprogramming factor Klf4 in the activation of the MET expression program. Using ChIP-Seq I mapped the binding of Klf4 and BMP Smads across the genome and linked their co-binding to a MET expression program determined by RNA-Seq. My work uncovers a thus-far unreported interaction between Klf4 and BMP signaling in cellular epithelialization that can directly improve the technical methods of reprogramming and have important implications for the induction of epithelial tissues in general.

Growth of bacteria in beer leads to turbidity and off-flavors, resulting in a spoiled and unpalatable product and thus economic loss. The most common beer-spoilage organisms (BSOs) are lactic acid bacteria (LAB), with Lactobacillus and Pediococcus species being the most problematic. Because of the harsh environment (low nutrients, antimicrobial compounds ethanol and hops, anaerobic), only select isolates are able to sustain growth in and spoil beer. To begin understanding the phenomenon of LAB adapting to overcome stresses in beer, ethanol tolerance, hop resistance, and nutrient acquisition mechanisms were investigated. First, ethanol tolerance was analyzed in the context of beer-spoilage ability, and it was found that it is intrinsically high in LAB, thus leading to the conclusion that LAB ability to spoil beer is not dependent on ethanol resistance levels. This was then followed by genome sequencing of the BSO Pediococcus claussenii ATCC BAA-344T (Pc344) to elucidate mechanisms being used to resist hops and acquire low abundance or alternative nutrients. Subsequent analysis of Pc344 and Lactobacillus brevis BSO 464 via reverse transcription quantitative PCR demonstrated the variability found among BSOs in the presence of beer-spoilage-related genes and their use during growth in beer. Further analysis of Pc344 was performed via RNA-sequencing to get a global view of gene expression during mid-logarithmic growth in beer. It was found that several alternative nutrients were being used by Pc344 to sustain growth, and that hop resistance was enabled by a variety of mechanisms including oxidative stress response and pH control. Finally, genomic comparison of BSOs determined that conservation is only present for closely related organisms and that no specific genes/proteins are indicative of an isolate’s beer-spoilage potential. It is more likely that horizontal gene transfer plays a major role in LAB adaption for growth in beer, and that plasmids are very important for this evolution, as was demonstrated by plasmid-variants of Pc344. The main conclusions of this thesis are therefore that hop resistance is the main factor determining ability to grow in beer, and that transfer of genetic elements is the driving force behind LAB evolving into BSOs.