The entire history of HIV-1 is hidden in its ten thousand bases, where information regarding its evolutionary traversal through the human population can only be unlocked with fine-scale sequence analysis. Measurable footprints of mutation and recombination have imparted upon us a wealth of knowledge, from multiple chimpanzee-to-human transmissions to patterns of neutralizing antibody and drug resistance. Extracting maximum understanding from such diverse data can only be accomplished by analyzing the viral population from many angles. This body of work explores two primary aspects of HIV sequence evolution, point mutation and recombination, through cross-sectional (inter-individual) and longitudinal (intra-individual) investigations, respectively. Cross-sectional Analysis: The role of Haiti in the subtype B pandemic has been hotly debated for years; while there have been many studies, up to this point, no one has incorporated the well-known mechanism of retroviral recombination into their biological model. Prior to the use of recombination detection, multiple analyses produced trees where subtype B appears to have first entered Haiti, followed by a jump into the rest of the world. The results presented here contest the Haiti-first theory of the pandemic and instead suggest simultaneous entries of subtype B into Haiti and the rest of the world. Longitudinal Analysis: Potential N-linked glycosylation sites (PNGS) are the most evolutionarily dynamic component of one of the most evolutionarily dynamic proteins known to date. While the number of mutations associated with the increase or decrease of PNGS frequency over time is high, there are a set of relatively stable sites that persist within and between longitudinally sampled individuals. Here, I identify the most conserved stable PNGSs and suggest their potential roles in host-virus interplay. In addition, I have identified, for the first time, what may be a gp-120-based environmental preference for N-linked glycosylation sites.

Sparsity has become an important modeling tool in areas such as genetics, signal and audio processing, medical image processing, etc. Via the penalization of l-1 norm based regularization, the structured sparse learning algorithms can produce highly accurate models while imposing various predefined structures on the data, such as feature groups or graphs. In this thesis, I first propose to solve a sparse learning model with a general group structure, where the predefined groups may overlap with each other. Then, I present three real world applications which can benefit from the group structured sparse learning technique. In the first application, I study the Alzheimer's Disease diagnosis problem using multi-modality neuroimaging data. In this dataset, not every subject has all data sources available, exhibiting an unique and challenging block-wise missing pattern. In the second application, I study the automatic annotation and retrieval of fruit-fly gene expression pattern images. Combined with the spatial information, sparse learning techniques can be used to construct effective representation of the expression images. In the third application, I present a new computational approach to annotate developmental stage for Drosophila embryos in the gene expression images. In addition, it provides a stage score that enables one to more finely annotate each embryo so that they are divided into early and late periods of development within standard stage demarcations. Stage scores help us to illuminate global gene activities and changes much better, and more refined stage annotations improve our ability to better interpret results when expression pattern matches are discovered between genes.

We propose a novel solution to prevent cancer by developing a prophylactic cancer. Several sources of antigens for cancer vaccines have been published. Among these, antigens that contain a frame-shift (FS) peptide or viral peptide are quite attractive for a variety of reasons. FS sequences, from either mistake in RNA processing or in genomic DNA, may lead to generation of neo-peptides that are foreign to the immune system. Viral peptides presumably would originate from exogenous but integrated viral nucleic acid sequences. Both are non-self, therefore lessen concerns about development of autoimmunity. I have developed a bioinformatical approach to identify these aberrant transcripts in the cancer transcriptome. Their suitability for use in a vaccine is evaluated by establishing their frequencies and predicting possible epitopes along with their population coverage according to the prevalence of major histocompatibility complex (MHC) types. Viral transcripts and transcripts with FS mutations from gene fusion, insertion/deletion at coding microsatellite DNA, and alternative splicing were identified in NCBI Expressed Sequence Tag (EST) database. 48 FS chimeric transcripts were validated in 50 breast cell lines and 68 primary breast tumor samples with their frequencies from 4% to 98% by RT-PCR and sequencing confirmation. These 48 FS peptides, if translated and presented, could be used to protect more than 90% of the population in Northern America based on the prediction of epitopes derived from them. Furthermore, we synthesized 150 peptides that correspond to FS and viral peptides that we predicted would exist in tumor patients and we tested over 200 different cancer patient sera. We found a number of serological reactive peptide sequences in cancer patients that had little to no reactivity in healthy controls; strong support for the strength of our bioinformatic approach. This study describes a process used to identify aberrant transcripts that lead to a new source of antigens that can be tested and used in a prophylactic cancer vaccine. The vast amount of transcriptome data of various cancers from the Cancer Genome Atlas (TCGA) project will enhance our ability to further select better cancer antigen candidates.

Proteins are a fundamental unit in biology. Although proteins have been extensively studied, there is still much to investigate. The mechanism by which proteins fold into their native state, how evolution shapes structural dynamics, and the dynamic mechanisms of many diseases are not well understood. In this thesis, protein folding is explored using a multi-scale modeling method including (i) geometric constraint based simulations that efficiently search for native like topologies and (ii) reservoir replica exchange molecular dynamics, which identify the low free energy structures and refines these structures toward the native conformation. A test set of eight proteins and three ancestral steroid receptor proteins are folded to 2.7Å all-atom RMSD from their experimental crystal structures. Protein evolution and disease associated mutations (DAMs) are most commonly studied by in silico multiple sequence alignment methods. Here, however, the structural dynamics are incorporated to give insight into the evolution of three ancestral proteins and the mechanism of several diseases in human ferritin protein. The differences in conformational dynamics of these evolutionary related, functionally diverged ancestral steroid receptor proteins are investigated by obtaining the most collective motion through essential dynamics. Strikingly, this analysis shows that evolutionary diverged proteins of the same family do not share the same dynamic subspace. Rather, those sharing the same function are simultaneously clustered together and distant from those functionally diverged homologs. This dynamics analysis also identifies 77% of mutations (functional and permissive) necessary to evolve new function. In silico methods for prediction of DAMs rely on differences in evolution rate due to purifying selection and therefore the accuracy of DAM prediction decreases at fast and slow evolvable sites. Here, we investigate structural dynamics through computing the contribution of each residue to the biologically relevant fluctuations and from this define a metric: the dynamic stability index (DSI). Using DSI we study the mechanism for three diseases observed in the human ferritin protein. The T30I and R40G DAMs show a loss of dynamic stability at the C-terminus helix and nearby regulatory loop, agreeing with experimental results implicating the same regulatory loop as a cause in cataracts syndrome.

Multi-task learning (MTL) aims to improve the generalization performance (of the resulting classifiers) by learning multiple related tasks simultaneously. Specifically, MTL exploits the intrinsic task relatedness, based on which the informative domain knowledge from each task can be shared across multiple tasks and thus facilitate the individual task learning. It is particularly desirable to share the domain knowledge (among the tasks) when there are a number of related tasks but only limited training data is available for each task. Modeling the relationship of multiple tasks is critical to the generalization performance of the MTL algorithms. In this dissertation, I propose a series of MTL approaches which assume that multiple tasks are intrinsically related via a shared low-dimensional feature space. The proposed MTL approaches are developed to deal with different scenarios and settings; they are respectively formulated as mathematical optimization problems of minimizing the empirical loss regularized by different structures. For all proposed MTL formulations, I develop the associated optimization algorithms to find their globally optimal solution efficiently. I also conduct theoretical analysis for certain MTL approaches by deriving the globally optimal solution recovery condition and the performance bound. To demonstrate the practical performance, I apply the proposed MTL approaches on different real-world applications: (1) Automated annotation of the Drosophila gene expression pattern images; (2) Categorization of the Yahoo web pages. Our experimental results demonstrate the efficiency and effectiveness of the proposed algorithms.