We present a cancer genomic analysis pipeline which takes

sequencing reads for both germline and tumor genomes as input and

outputs prioritized lists of the most affected genes in the tumor

genome. Using publicly available datasets and literature specific

to each patient, we extract out clinically relevant information

to be used in a novel reporting and ranking system in order to

identify the most affected genes and pathways within a

patient. Network-based approaches that integrate protein-protein,

protein-TF, and protein-drug interaction data are used to

identify potentially therapeutic drugs and their targets. Effects

of genetic variations on gene expression, as profiled by RNA-seq

in tumor samples, are used to provide further evidence of

“driver” mutations -- those mutations responsible for tumor

progression. Additionally, previously implicated small and large

variations (including gene fusions) are reported.

Results are presented in a collaborative interface that combines

all evidence for the top ranking genes and pathways. Affected

genes in and around protein coding sequences are investigated

further using sequence-level features such as predicted secondary

structure, solvent accessibility, phosphorylation status, and

protein domains. By using an integrative approach, effects of

genetic variations on gene expression are used to provide further

evidence of “driver” mutations.

This pipeline has been developed with the aim to be used in

assisting in the analysis of pediatric tumors, as an unbiased and

automated method. We present results that agree with previous

literature and highlight specific findings in a few

patients. Portions of this pipeline have been successfully reused

in the analysis of other high-throughput sequencing data in

non-cancer related projects. This work provides a basis for which

future personalized medicine pipelines can be systematically

performed in order to assist in the treatment of newly diagnosed

cancer patients.

High-throughput sequencing (HTS) technologies play important roles in the life sciences by allowing the rapid parallel sequencing of very large numbers of relatively short nucleotide sequences, in applications ranging from genome sequencing and resequencing to digital microarrays and ChIP-Seq experiments. As experiments scale up, HTS technologies create new bioinformatics challenges for the storage, distribution, and downstream analyses of HTS data. In this thesis, we address the challenges and opportunities in the storage and downstream analyses of data derived from HTS technologies.

To address the growing amount of HTS data, we develop data structures and compression algorithms for HTS data storage. A processing stage maps short sequences to a reference genome or a large table of sequences. Then the integers representing the short sequence absolute or relative addresses, their length, and the substitutions they may contain are compressed and stored using various entropy coding algorithms, including both old and new fixed codes (e.g Golomb, Elias Gamma, MOV) and variable codes (e.g. Huffman). The general methodology is illustrated and applied to several HTS data sets. Results show that the information contained in HTS files can be compressed by a factor of 10 or more, depending on the statistical properties of the data sets and various other choices and constraints. Our algorithms fair well against general purpose compression programs such as gzip, bzip2 and 7zip; timing results show that our algorithms are consistently faster than the best general purpose compression programs.

Achieving a comprehensive map of all the regulatory elements encoded in the human genome is a fundamental challenge of biomedical research. So far, only a small fraction of the regulatory elements have been characterized, and there is great interest in applying computational techniques to systematically discover these elements. Such efforts, however, have been significantly hindered by the overwhelming size of non-coding DNA regions and the statistical variability and complex spatial organizations of mammalian regulatory elements. The MotifMap system uses databases of transcription factor binding motifs, refined genome alignments, and a comparative genomic statistical approach to provide comprehensive maps of candidate regulatory elements encoded in the genomes of model species.

MotifMap and its integration with other data provide a foundation for analyzing gene reg- ulation on a genome-wide scale, and for automatically generating regulatory pathways and hypotheses. The power of this approach is demonstrated and discussed using the P53 apop- totic pathway and the Gli hedgehog pathways as examples.

Further application of MotifMap is underlined by its integration into CircadiOmics (de- veloped in the Baldi lab), which aims to decode transcriptional machinery that are under circadian control. Here we utilize MotifMap in understanding and delineating the roles of a subset of the sirtuin family of deacetylases in regulating circadian rhythms. Circadian rhythms are intimately linked to cellular metabolism. Specifically, the NAD+-dependent deacetylase SIRT1, the founding member of the sirtuin family, contributes to clock func- tion. Whereas SIRT1 exhibits diversity in deacetylation targets and subcellular localization, SIRT6 is the only constitutively chromatin-associated sirtuin and is prominently present at transcriptionally active genomic loci. Comparison of the hepatic circadian transcriptomes reveals that SIRT6 and SIRT1 separately control transcriptional specificity and therefore define distinctly partitioned classes of circadian genes. SIRT6 interacts with CLOCK:BMAL1 and, differently from SIRT1, governs their chromatin recruitment to circadian gene promot- ers. Moreover, SIRT6 controls circadian chromatin recruitment of SREBP-1, resulting in the cyclic regulation of genes implicated in fatty acid and cholesterol metabolism. This mechanism parallels a phenotypic disruption in fatty acid metabolism in SIRT6 null mice as revealed by circadian metabolome analyses. Thus, genomic partitioning by two independent sirtuins contributes to differential control of circadian metabolism.

The study of circadian rhythms increasingly relies on high throughput circadian omic data. Processing, annotating and analyzing these data require an efficient and powerful bioinformatics software system. To this end, we have implemented such a system. Major components of this system include genome-wide binding sites predictors, statistical and machine learning pipelines, functional enrichment pipelines, visualization tools and a data management server. Using this system, we have produced high impact publications, a circadian web portal (CircadiOmics, http://circadiomics.ics.uci.edu/), and advanced our understanding of the organization of the circadian transcriptome.

Achieving human-level performance at predicting chemical reactions remains an open prob- lem with broad potential applications. Here we describe a deep learning-based tool for chemical reaction prediction and product identification. Significant efforts were made to curate and refine a new, high-quality data set of hand-selected chemical reactions written at the level of elementary electron movements. Using deep artificial neural networks trained on this data, we demonstrate a high degree of accuracy at predicting real-world reactions. Because predictions are made at the elementary step level, they can be chained together to form multi-step reaction pathway searches, to help identify unknown side products.

We also present a computational brewing application, COBRA, capable of simulating com- plex chemical mixtures. We demonstrate its efficacy at modeling both the photooxidation of isoprene, and the oxidation of squalene in the presence of ozone, by comparing predicted results with results obtained from high-resolution mass spectrometry.

In addition, we address the problem of atom-mapping for chemical reactions, by designing a new atom-mapping algorithm that can be used to annotate unmapped reactions.

Data from decades of life sciences research and literature is being curated and made available for searching and analysis. While considerable work has been done to integrate and re-use this data, what is still lacking is a unifying platform that allows new experimental data to leverage all previously published data effectively.

Crick is an intelligent and scalable platform for data integration, visualization, and searching for meaningful biological hypothesis. It was built to create an effective way to integrate and analyze experimental data in the context of the vast literature of other biologically relevant information. Crick was designed ground-up to solve some of the most challenging problems in biological data such as entity resolution, size, scale, reliability of the data, visualization of high-dimensional information, etc. Crick has been successfully used to identify molecular mechanism regulating circadian metabolism; to understand the complex coupling of circadian oscillating species; to study pediatric cancers; to analyze the dynamic long-range interactions in the genome; and more.

Circadian oscillations play a fundamental role in many biological processes including cell metabolism and cell cycle. As such, interest in understanding these molecular oscillations has generated an increasingly large collection of circadian omic data. These studies have demonstrated the remarkable plasticity with which the set of oscillating molecular species within a cell are selected. These large shifts in oscillating species are known as circadian reprogramming events. These events have been observed across experimental condition, tissue, and species. While many of these studies have made tissue or condition specific conclusions, a consolidated framework of software tools and a central repository of data has become necessary to answer questions about the orchestration of these reprogramming events.

Deep learning has recently achieved many breakthroughs. Neural networks - the models behind deep learning - have a large number of hyperparameters whose correct settings are crucial to obtain optimal performance. As the need for hyperparameter optimization has grown, much research has been produced on this topic.

Existing hyperparameter optimization methods often assume a noiseless objective function. In this thesis we explore the effect of noise in the objective function, that is the neural network training, on the hyperparameter optimization. The thesis begins by motivating hyperparameter search through an applied machine learning problem in the domain of neutrino physics. With the help of hyperparameter optimization we develop an energy estimator for observations from the NOvA experiment at Fermilab. This energy estimator is able to outperform prior methods in terms of prediction accuracy by using a deep neural network to directly predict the target energy from the detector response. We then introduce hyperparameter optimization software, \textit{Sherpa}, that has been developed as part of this thesis. After that we focus on methods for the selection of optimal hyperparameter settings when observations are noisy. This is solved through a group sequential testing framework that results in an equivalence class of hyperparameter configurations. The method is empirically validated on three machine learning tasks and is shown to return the optimal hyperparameter setting at a higher rate than choosing the best observed. It furthermore increases reproducibility by reducing variance in the outcome of the search. Lastly, we focus on finding optimal trade-offs between repeated evaluation of hyperparameter settings and exploration of the space. In particular, we benchmark a number of popular hyperparameter search methods on machine learning tasks with high noise between runs. Empirical results show that standard Gaussian process based Bayesian optimization without repetition tends to deliver competitive results at even small computational budgets in high noise hyperparameter optimization. The thesis concludes with suggestions for future work.

Circadian oscillations play a fundamental role in many biological processes including cell metabolism and cell cycle. As such, interest in understanding these molecular oscillations has generated an increasingly large collection of circadian omic data. These studies have demonstrated the remarkable plasticity with which the set of oscillating molecular species within a cell are selected. These large shifts in oscillating species are known as circadian reprogramming events. These events have been observed across experimental condition, tissue, and species. While many of these studies have made tissue or condition specific conclusions, a consolidated framework of software tools and a central repository of data has become necessary to answer questions about the orchestration of these reprogramming events.

By combining the largest repository for circadian omic data with improved methods for the detection of circadian oscillation and regulation in omic studies, a model for the transcriptomic organization of circadian rhythms can be identified. Results are presented from a collection of specific reprogramming studies utilizing these new methods, as well as a high through-put analysis of a large collection of comparable transcriptomic datasets from mouse tissue. The identified model can be viewed as a deep hierarchical network of circadian regulation that originates from the core circadian clock to over 95% of oscillating transcripts.