Biocomputing 2011

The following sections are included:

• Preface

• Acknowledgments

The following sections are included:

• Introduction

• Session Summary

• Acknowledgments

Many methods have been proposed for facilitating the uncovering of genes that underlie the pathology of different diseases. Some are purely statistical, resulting in a (mostly) undifferentiated set of genes that are differentially expressed (or co-expressed), while others seek to prioritize the resulting set of genes through comparison against specific known targets. Most of the recent approaches use either single data or knowledge sources, or combine the independent predictions from each source. However, given that multiple kinds of heterogeneous sources are potentially relevant for gene prioritization, each subject to different levels of noise and of varying reliability, each source bearing information not carried by another, we claim that an ideal prioritization method should provide ways to discern amongst them in a true integrative fashion that captures the subtleties of each, rather than using a simple combination of sources. Integration of multiple data for gene prioritization is thus more challenging than its single data type counterpart. What we propose is a novel, general, and flexible formulation that enables multi-source data integration for gene prioritization that maximizes the complementary nature of different data and knowledge sources in order to make the most use of the information content of aggregate data. Protein-protein interactions and Gene Ontology annotations were used as knowledge sources, together with assay-specific gene expression and genome-wide association data. Leave-one-out testing was performed using a known set of Alzheimer's Disease genes to validate our proposed method. We show that our proposed method performs better than the best multi-source gene prioritization systems currently published.

The precise molecular etiology of obstructive sleep apnea (OSA) is unknown; however recent research indicates that several interconnected aberrant pathways and molecular abnormalities are contributors to OSA. Identifying the genes and pathways associated with OSA can help to expand our understanding of the risk factors for the disease as well as provide new avenues for potential treatment. Towards these goals, we have integrated relevant high dimensional data from various sources, such as genome-wide expression data (microarray), protein-protein interaction (PPI) data and results from genome-wide association studies (GWAS) in order to define sub-network elements that connect some of the known pathways related to the disease as well as define novel regulatory modules related to OSA. Two distinct approaches are applied to identify sub-networks significantly associated with OSA. In the first case we used a biased approach based on sixty genes/proteins with known associations with sleep disorders and/or metabolic disease to seed a search using commercial software to discover networks associated with disease followed by information theoretic (mutual information) scoring of the sub-networks. In the second case we used an unbiased approach and generated an interactome constructed from publicly available gene expression profiles and PPI databases, followed by scoring of the network with p-values from GWAS data derived from OSA patients to uncover sub-networks significant for the disease phenotype. A comparison of the approaches reveals a number of proteins that have been previously known to be associated with OSA or sleep. In addition, our results indicate a novel association of Phosphoinositide 3-kinase, the STAT family of proteins and its related pathways with OSA.

General pedigrees can be encoded as Bayesian networks, where the common MPE query corresponds to finding the most likely haplotype configuration. Based on this, a strategy for grid parallelization of a state-of-the-art Branch and Bound algorithm for MPE is introduced: independent worker nodes concurrently solve subproblems, managed by a Branch and Bound master node. The likelihood functions are used to predict subproblem complexity, enabling efficient automation of the parallelization process. Experimental evaluation on up to 20 parallel nodes yields very promising results and suggest the effectiveness of the scheme, solving several very hard problem instances. The system runs on loosely coupled commodity hardware, simplifying deployment on a larger scale in the future.

While networks models have often been applied to complex biological systems, they are increasingly being implemented to investigate clinical questions. Clinical trials have been studied extensively by traditional statistical methods but never, to our knowledge, using networks. We obtained data for 6,847 clinical trials from five "Nervous System Diseases" (NSD) and five "Behaviors and Mental Disorders" (BMD) from the clinicaltrials.gov registry. We constructed networks of diseases and interventions for visualization and analysis using Cytoscape software. To standardize nomenclature and enable multi-level annotation, we used MeSH and UMLS terms. We then constructed separate BMD and NSD networks to study dynamics over time. To assess how topology features related to clinical significance, we constructed a sub-network of Multiple Sclerosis and Alzheimer's trials and identified which trials had been published in high-profile medical journals. We found that the BMD network has evolved into a large, decentralized topology and does not distinctly reflect the five diseases by which it was defined, while the NSD network does, though other diseases and sub-phenotypes have emerged as areas of research. We also found that high-profile trials have distinctive network characteristics. Future work is needed to address mathematical questions such as scale-dependence of network features, clinical questions such as trial design optimization, and methodological questions such as data quality improvement.

Gene set analyses have become a standard approach for increasing the sensitivity of transcriptomic studies. However, analytical methods incorporating gene sets require the availability of pre-defined gene sets relevant to the underlying physiology being studied. For novel physiological problems, relevant gene sets may be unavailable or existing gene set databases may bias the results towards only the best-studied of the relevant biological processes. We describe a successful attempt to mine novel functional gene sets for translational projects where the underlying physiology is not necessarily well characterized in existing annotation databases. We choose targeted training data from public expression data repositories and define new criteria for selecting biclusters to serve as candidate gene sets. Many of the discovered gene sets show little or no enrichment for informative Gene Ontology terms or other functional annotation. However, we observe that such gene sets show coherent differential expression in new clinical test data sets, even if derived from different species, tissues, and disease states. We demonstrate the efficacy of this method on a human metabolic data set, where we discover novel, uncharacterized gene sets that are diagnostic of diabetes, and on additional data sets related to neuronal processes and human development. Our results suggest that our approach may be an efficient way to generate a collection of gene sets relevant to the analysis of data for novel clinical applications where existing functional annotation is relatively incomplete.

RNA virus phenotypic changes often result from multiple alternative molecular mechanisms, where each mechanism involves changes to a small number of key residues. Accordingly, we propose to learn genotype-phenotype functions, using Disjunctive Normal Form (DNF) as the assumed functional form. In this study we develop DNF learning algorithms that attempt to construct predictors as Boolean combinations of covariates. We demonstrate the learning algorithm's consistency and efficiency on simulated sequences, and establish their biological relevance using a variety of real RNA virus datasets representing different viral phenotypes, including drug resistance, antigenicity, and pathogenicity. We compare our algorithms with previously published machine learning algorithms in terms of prediction quality: leave-one-out performance shows superior accuracy to other machine learning algorithms on the HIV drug resistance dataset and the UCIs promoter gene dataset. The algorithms are powerful in inferring the genotype-phenotype mapping from a moderate number of labeled sequences, as are typically produced in mutagenesis experiments. They can also greedily learn DNFs from large datasets. The Java implementation of our algorithms will be made publicly available.

Genome-wide associations studies (GWAS) have been very successful in identifying common genetic variation associated to numerous complex diseases [1]. However, most of the identified common genetic variants appear to confer modest risk and few causal alleles have been identified [2]. Furthermore, these associations account for a small portion of the total heritability of inherited disease variation [1]. This has led to the reexamination of the contribution of environment, gene-gene and gene-environment interactions, and rare genetic variants in complex diseases [1, 3, 4]. There is strong evidence that rare variants play an important role in complex disease etiology and may have larger genetic effects than common variants [2].

Currently, much of what we know regarding the contribution of rare genetic variants to disease risk is based on a limited number of phenotypes and candidate genes. However, rapid advancement of second generation sequencing technologies will invariably lead to widespread association studies comparing whole exome and eventually whole genome sequencing of cases and controls. A tremendous challenge for enabling these "next generation" medical genomic studies is developing statistical approaches for correlating rare genetic variants with disease outcome.

The analysis of rare variants is challenging since methods used for common variants are woefully underpowered. Therefore, methods that can deal with genetic heterogeneity at the trait-associated locus have been developed to analyze rare variants. These methods instead analyzing individual variants analyze variants within a region/gene as a group and usually rely on collapsing. They can be applied to both in cases vs. controls and quantitative trait studies are needed. The paper of Bansal et al. in this volume describes the application of a number of statistical methods for testing associations between rare variants in two genes to obesity. The authors considered the relative merits of the different methods as well as important implementation details, such as the leveraging of genomic annotations and determining p-values.

Knowledge of haplotypes can increase the power of GWAS studies and also highlight associations that are impossible to detect without haplotype phase (e.g. loss of heterozygosity). Even more complicated phase-dependent interactions of variants in linkage equilibrium have also been suggested as possible causes of missing heritability. In their work, Hallsorsson et al. formulate algorithmic strategies for haplotype phasing by multi-assembly of shared haplotypes from next-generation sequencing data. These methods would allow testing haplotypes harboring rare variants for association and potentially increase their explanatory power.

Since single SNP tests are often underpowered in rare variant association analysis, Zeggini and Asimit propose a locus-based method that has high power in the presence of rare variants and that incorporate base quality scores available for sequencing data. Their results suggest that this multi-marker approach may be best suited for smaller regions, or after some filtering to reduce the number of SNPs that are jointly tested to reduce loss of power due to multiple-testing adjustments.

Finally, the paper of Zhou et al., presents a penalized regression framework for association testing on sequence data, in the presence of both common and rare variants. This method also introduces the use of weights to incorporate available biological information on the variants. Although these tactics improve both false positive and false negative rates, they represent an incremental development and there is still significant room for improvement.

With the development of sequencing technologies and methods to detect complex trait rare variant associations many new and exciting discovery are imminent. The analysis of rare variants is still in its infancy and the next few years promises to produce many new methods to meet the special demands of analyzing this type of data.

Note from Publisher: This article contains the abstract and references.

The contribution of collections of rare sequence variations (or 'variants') to phenotypic expression has begun to receive considerable attention within the biomedical research community. However, the best way to capture the effects of rare variants in relevant statistical analysis models is an open question. In this paper we describe the application of a number of statistical methods for testing associations between rare variants in two genes to obesity. We consider the relative merits of the different methods as well as important implementation details, such as the leveraging of genomic annotations and determining p-values.

In this paper we propose algorithmic strategies, Lander-Waterman-like statistical estimates, and genome-wide software for haplotype phasing by multi-assembly of shared haplotypes. Specifically, we consider four types of results which together provide a comprehensive workflow of GWAS data sets: (1) statistics of multi-assembly of shared haplotypes (2) graph theoretic algorithms for haplotype assembly based on conflict graphs of sequencing reads (3) inference of pedigree structure through haplotype sharing via tract finding algorithms and (4) multi-assembly of shared haplotypes of cases, controls, and trios. The input for the workflows that we consider are any of the combination of: (A) genotype data (B) next generation sequencing (NGS) (C) pedigree information.

(1) We present Lander-Waterman-like statistics for NGS projects for the multi-assembly of shared haplotypes. Results are presented in Sec. 2. (2) In Sec. 3, we present algorithmic strategies for haplotype assembly using NGS, NGS + genotype data, and NGS + pedigree information. (3) This work builds on algorithms presented in Halldórsson et al.¹ and are part of the same library of tools co-developed for GWAS workflows. (4) Section 3.3.1 contains algorithmic strategies for multi-assembly of GWAS data. We present algorithms for assembling large data sets and for determining and using shared haplotypes to more reliably assemble and phase the data. Workflows 1-4 provide a set of rigorous algorithms which have the potential to identify phase-dependent interactions between rare variants in linkage equilibrium which are associated with cases. They build on our extensive work on haplotype phasing,^1–3 haplotype assembly,^4,5 and whole genome assembly comparison.⁶

There is growing interest in the role of rare variants in multifactorial disease etiology, and increasing evidence that rare variants are associated with complex traits. Single SNP tests are underpowered in rare variant association analyses, so locus-based tests must be used. Quality scores at both the SNP and genotype level are available for sequencing data and they are rarely accounted for. A locus-based method that has high power in the presence of rare variants is extended to incorporate such quality scores as weights, and its power is compared with the original method via a simulation study. Preliminary results suggest that taking uncertainty into account does not improve the power.

Whole exome and whole genome sequencing are likely to be potent tools in the study of common diseases and complex traits. Despite this promise, some very difficult issues in data management and statistical analysis must be squarely faced. The number of rare variants identified by sequencing is apt to be much larger than the number of common variants encountered in current association studies. The low frequencies of rare variants alone will make association testing difficult. This article extends the penalized regression framework for model selection in genome-wide association data to sequencing data with both common and rare variants. Previous research has shown that lasso penalties discourage irrelevant predictors from entering a model. The Euclidean penalties dealt with here group variants by gene or pathway. Pertinent biological information can be incorporated by calibrating penalties by weights. The current paper examines some of the tradeoffs in using pure lasso penalties, pure group penalties, and mixtures of the two types of penalty. All of the computational and statistical advantages of lasso penalized estimation are retained in this richer setting. The overall strategy is implemented in the free statistical genetics analysis software MENDEL and illustrated on both simulated and real data.

Recent advances in sequencing technologies have made is possible, for the first time, to take a thorough census of the microbial species present in a given environment. This presents a particularly exciting opportunity since bacteria and archea comprise the dominant forms of life on earth, and since they are vital to human health and to the wellbeing of our environment. However, the bioinformatics for interpreting these very large sequence datasets are not fully developed. This session presents recent work supporting the computational analysis of microbiome data.

In many situations we are faced with the need to estimate the number of classes in a population from observed count data: this arises not only in biology, where we are interested in the number of taxa such as species, but also in many other fields such as public health, criminal justice, software engineering, etc. This problem has a rich history in theoretical statistics, dating back at least to 1943, and many approaches have been proposed and studied. However, to date only one approach has been implemented in readily available software, namely a relatively simple nonparametric method which, while straightforward to program, is not flexible and can be prone to information loss. Here we present CatchAll, a new, platform-independent, user-friendly, computationally optimized software package which calculates a powerful and flexible suite of parametric models (based on current statistical research) in addition to all existing nonparametric procedures. We briefly describe the software and its mathematical underpinnings (which are treated in depth elsewhere), and we work through an applied example from microbial ecology in detail.

The human body is home to a diverse assemblage of microbial species. In fact, the number of microbial cells in each person is an order of magnitude greater than the number of cells that make up the body itself. Changes in the composition and relative abundance of these microbial species are highly associated with intestinal and respiratory disorders and diseases of the skin and mucus membranes. While cultivation-independent methods employing PCR-amplification, cloning and sequence analysis of 16S rRNA or other phylogenetically informative genes have made it possible to assess the composition of microbial species in natural environments, until recently this approach has been too time consuming and expensive for routine use. Advances in high throughput pyrosequencing have largely eliminated these obstacles, reducing cost and increasing sequencing capacity by orders of magnitude. In fact, although numerous arithmetic and statistical measurements are available to assess the composition and diversity of microbial communities, the limiting factor has become applying these analyses to millions of sequences and visualizing the results. We introduce a new, easy-to-use, extensible visualization and analysis software framework that facilitates the manipulation and interpretation of large amounts of metagenomic sequence data. The framework automatically performs an array of standard metagenomic analyses using FASTA files that contain 16S rRNA sequences as input. The framework has been used to reveal differences between the composition of the microbiota in healthy individuals and individuals with diseases such as bacterial vaginosis and necrotizing enterocolitis.

This article explains the statistical and computational methodology used to analyze species abundances collected using the LNBL Phylochip in a study of Irritable Bowel Syndrome (IBS) in rats.

Some tools already available for the analysis of ordinary microarray data are useful in this type of statistical analysis. For instance in correcting for multiple testing we use Family Wise Error rate control and step-down tests (available in the multtest package). Once the most significant species are chosen we use the hypergeometric tests familiar for testing GO categories to test specific phyla and families.

We provide examples of normalization, multivariate projections, batch effect detection and integration of phylogenetic covariation, as well as tree equalization and robustification methods.

High-throughput sequencing technology has opened the door to the study of the human microbiome and its relationship with health and disease. This is both an opportunity and a significant biocomputing challenge. We present here a 3D visualization methodology and freely-available software package for facilitating the exploration and analysis of high-dimensional human microbiome data. Our visualization approach harnesses the power of commercial video game development engines to provide an interactive medium in the form of a 3D heat map for exploration of microbial species and their relative abundance in different patients. The advantage of this approach is that the third dimension provides additional layers of information that cannot be visualized using a traditional 2D heat map. We demonstrate the usefulness of this visualization approach using microbiome data collected from a sample of premature babies with and without sepsis.

16S rRNA gene sequencing has been widely used for probing the species structure of a variety of environmental bacterial communities. Alternatively, 16S rRNA gene fragments can be retrieved from shotgun metagenomic sequences (metagenomes) and used for species profiling. Both approaches have their limitations—16S rRNA sequencing may be biased because of unequal amplification of species' 16S rRNA genes, whereas shotgun metagenomic sequencing may not be deep enough to detect the 16S rRNA genes of rare species in a community. However, previous studies showed that these two approaches give largely similar species profiles for a few bacterial communities. To investigate this problem in greater detail, we conducted a systematic comparison of these two approaches. We developed PHYLOSHOP, a pipeline that predicts 16S rRNA gene fragments in metagenomes, reports the taxonomic assignment of these fragments, and visualizes their taxonomy distribution. Using PHYLOSHOP, we analyzed 33 metagenomic datasets of human-associated bacterial communities, and compared the bacterial community structures derived from these metagenomic datasets with the community structure derived from 16S rRNA gene sequencing (71 datasets). Based on several statistical tests (including a statistical test proposed here that takes into consideration differences in sample size), we observed that these two approaches give significantly different community structures for nearly all the bacterial communities collected from different locations on and in human body, and that these differences cannot be be explained by differences in sample size and are likely to be attributed by experimental method.

The following sections are included:

• Background

• Session Summary

• Acknowledgements

The p38 MAP kinases play a critical role in regulating stress-activated pathways, and serve as molecular targets for controlling inflammatory diseases. Computer-aided efforts for developing p38 inhibitors have been hampered by the necessity to include the enzyme conformational flexibility in ligand docking simulations. A useful strategy in such complicated cases is to perform ensemble-docking provided that a representative set of conformers is available for the target protein either from computations or experiments. We explore here the abilities of two computational approaches, molecular dynamics (MD) simulations and anisotropic network model (ANM) normal mode analysis, for generating potential ligand-bound conformers starting from the apo state of p38, and benchmark them against the space of conformers (or the reference modes of structural changes) inferred from principal component analysis of 134 experimentally resolved p38 kinase structures. ANM-generated conformations are found to provide a significantly better coverage of the inhibitor-bound conformational space observed experimentally, compared to MD simulations performed in explicit water, suggesting that ANM-based sampling of conformations can be advantageously employed as input structural models in docking simulations.

Collagen is a ubiquitous extracellular matrix protein. Its biological functions, including maintenance of the structural integrity of tissues, depend on its multiscale, hierarchical structure. Three elongated, twisted peptide chains of > 1000 amino acids each assemble into trimeric proteins characterized by the defining triple helical domain. The trimers associate into fibrils, which pack into fibers. We conducted a 10 ns molecular dynamics simulation of the full-length triple helical domain, which was made computationally feasible by segmenting the protein into overlapping fragments. The calculation included ~1.8 million atoms, including solvent, and took approximately 11 months using the CPUs of over a quarter of a million computers. Specialized analysis protocols and a relational database were developed to process the large amounts of data, which are publicly available. The simulated structures exhibit heterogeneity in the triple helical domain consistent with experimental results but at higher resolution. The structures serve as the foundation for studies of higher order forms of the protein and for modeling the effects of disease-associated mutations.

Subsequent to the peptidyl transfer step of the translation elongation cycle, the initially formed pre-translocation ribosome, which we refer to here as R₁, undergoes a ratchet-like intersubunit rotation in order to sample a rotated conformation, referred to here as R_F, that is an obligatory intermediate in the translocation of tRNAs and mRNA through the ribosome during the translocation step of the translation elongation cycle. R_F and the R₁ to R_F transition are currently the subject of intense research, driven in part by the potential for developing novel antibiotics which trap R_F or confound the R₁ to R_F transition. Currently lacking a 3D atomic structure of the R_F endpoint of the transition, as well as a preliminary conformational trajectory connecting R₁ and R_F, the dynamics of the mechanistically crucial R₁ to R_F transition remain elusive. The current literature reports fitting of only a few ribosomal RNA (rRNA) and ribosomal protein (r-protein) components into cryogenic electron microscopy (cryo-EM) reconstructions of the Escherichia coli ribosome in RF. In this work we now fit the entire Thermus thermophilus 16S and 23S rRNAs and most of the remaining T. thermophilus r-proteins into a cryo-EM reconstruction of the E. coli ribosome in R_F in order to build an almost complete model of the T. thermophilus ribosome in R_F thus allowing a more detailed view of this crucial conformation. The resulting model validates key predictions from the published literature; in particular it recovers intersubunit bridges known to be maintained throughout the R₁ to R_F transition and results in new intersubunit bridges that are predicted to exist only in R_F. In addition, we use a recently reported E. coli ribosome structure, apparently trapped in an intermediate state along the R₁ to R_F transition pathway, referred to here as R₂, as a guide to generate a T. thermophilus ribosome in the R₂ state. This demonstrates a multiresolution method for morphing large complexes and provides us with a structural model of R₂ in the species of interest. The generated structural models form the basis for probing the motion of the deacylated tRNA bound at the peptidyl-tRNA binding site (P site) of the pre-translocation ribosome as it moves from its so-called classical P/P configuration to its so-called hybrid P/E configuration as part of the R₁ to R_F transition. We create a dynamic model of this process which provides structural insights into the functional significance of R₂ as well as detailed atomic information to guide the design of further experiments. The results suggest extensibility to other steps of protein synthesis as well as to spatially larger systems.

We have proposed a parallel simulated annealing using genetic crossover as one of powerful conformational search methods, in order to find the global minimum energy structures for protein systems. The simulated annealing using genetic crossover method, which incorporates the attractive features of the simulated annealing and the genetic algorithm, is useful for finding a minimum potential energy conformation of protein systems. However, when we perform simulations by using this method, we often find obviously unnatural stable conformations, which have "knots" of a string of an amino-acid sequence. Therefore, we combined knot theory with our simulated annealing using genetic crossover method in order to avoid the knot conformations from the conformational search space. We applied this improved method to protein G, which has 56 amino acids. As the result, we could perform the simulations, which avoid knot conformations.

The following sections are included:

• Introduction

• Session Summary

• Acknowledgments

Personal genome resequencing has provided promising lead to personalized medicine. However, due to the limited samples and the lack of case/control design, current interpretation of personal genome sequences has been mainly focused on the identification and functional annotation of the DNA variants that are different from the reference genome. The reference genome was deduced from a collection of DNAs from anonymous individuals, some of whom might be carriers of disease risk alleles. We queried the reference genome against a large high-quality disease-SNP association database and found 3,556 disease-susceptible variants, including 15 rare variants. We assessed the likelihood ratio for risk for the reference genome on 104 diseases and found high risk for type 1 diabetes (T1D) and hypertension. We further demonstrated that the risk of T1D was significantly higher in the reference genome than those in a healthy patient with a whole human genome sequence. We found that the high T1D risk was mainly driven by a R260W mutation in PTPN22 in the reference genome. Therefore, we recommend that the disease-susceptible variants in the reference genome should be taken into consideration and future genome sequences should be interpreted with curated and predicted disease-susceptible loci to assess personal disease risk.

The diagnosis and treatment of cancers, which rank among the leading causes of mortality in developed nations, presents substantial clinical challenges. The genetic and epigenetic heterogeneity of tumors can lead to differential response to therapy and gross disparities in patient outcomes, even for tumors originating from similar tissues. High-throughput DNA sequencing technologies hold promise to improve the diagnosis and treatment of cancers through efficient and economical profiling of complete tumor genomes, paving the way for approaches to personalized oncology that consider the unique genetic composition of the patient's tumor. Here we present a novel method to leverage the information provided by cancer genome sequencing to match an individual tumor genome with commercial cell lines, which might be leveraged as clinical surrogates to inform prognosis or therapeutic strategy. We evaluate the method using a published lung cancer genome and genetic profiles of commercial cancer cell lines. The results support the general plausibility of this matching approach, thereby offering a first step in translational bioinformatics approaches to personalized oncology using established cancer cell lines.

Personalized medicine is a high priority for the future of health care. The idea of tailoring an individual's wellness plan to their unique genetic code is one which we hope to realize through the use of pharmacogenomics. There have been examples of tremendous success in pharmacogenomic associations however there are many such examples in which only a small proportion of trait variance has been explained by the genetic variation. Although the increased use of GWAS could help explain more of this variation, it is likely that a significant proportion of the genetic architecture of these pharmacogenomic traits are due to complex genetic effects such as epistasis, also known as gene-gene interactions, as well as gene-drug interactions. In this study, we utilize the Biofilter software package to look for candidate epistasis contributing to risk for virologic failure with efavirenz-containing antiretroviral therapy (ART) regimens in treatment-naïve participants of AIDS Clinical Trials Group (ACTG) randomized clinical trials. A total of 904 individuals from three ACTG trials with data on efavirenz treatment are analyzed after race-stratification into white, black, and Hispanic ethnic groups. Biofilter was run considering 245 candidate ADME (absorption, distribution, metabolism, and excretion) genes and using database knowledge of gene and protein interaction networks to produce approximately 2 million SNP-SNP interaction models within each ethnic group. These models were evaluated within the PLATO software package using pair wise logistic regression models. Although no interaction model remained significant after correction for multiple comparisons, an interaction between SNPs in the TAP1 and ABCC9 genes was one of the top models before correction. The TAP1 protein is responsible for intracellular transport of antigen to MHC class I molecules, while ABCC9 codes for a transporter which is part of the subfamily of ABC transporters associated with multi-drug resistance. This study demonstrates the utility of the Biofilter method to prioritize the search for gene-gene interactions in large-scale genomic datasets, although replication in a larger cohort is required to confirm the validity of this particular TAP1-ABCC9 finding.

In this paper, we describe using Synthesis-View, a new method of presenting complex genetic data, to revisit results of a study from the BioVU Vanderbilt DNA databank. BioVU is a biorepository of DNA samples coupled with de-identified electronic medical records (EMR). In the Ritchie et al. study ¹ ~10,000 BioVU samples were genotyped for 21 SNPs that were previously associated with 5 diseases: atrial fibrillation, Crohn Disease, multiple sclerosis, rheumatoid arthritis, and type 2 diabetes. In the proof-of-concept study, the 21 tests of association replicated previous findings where sample size provided adequate power. The majority of the BioVU results were originally presented in tabular form. Herein we have revisited the results of this study using Synthesis-View. The Synthesis-View software tool visually synthesizes the results of complex, multi-layered studies that aim to characterize associations between small numbers of single-nucleotide polymorphisms (SNPs) and diseases and/or phenotypes, such as the results of replication and meta-analysis studies. Using Synthesis-View with the data of the Ritchie et al. study and presenting these data in this integrated visual format demonstrates new ways to investigate and interpret these kinds of data. Synthesis-View is freely available for non-commercial research institutions, for full details see https://chgr.mc.vanderbilt.edu/synthesisview.

Understanding how genetic variants impact the regulation and expression of genes is important for forging mechanistic links between variants and phenotypes in personal genomics studies. In this work, we investigate statistical interactions among variants that alter gene expression and identify 79 genes showing highly significant interaction effects consistent with genetic heterogeneity. Of the 79 genes, 28 have been linked to phenotypes through previous genomic studies. We characterize the structural and statistical nature of these 79 cis-epistasis models, and show that interacting regulatory SNPs often lie far apart from each other and can be quite distant from the gene they regulate. By using cis-epistasis models that account for more variance in gene expression, investigators may improve the power and replicability of their genomics studies, and more accurately estimate an individual's gene expression level, improving phenotype prediction.

High-throughput sequencing is currently a major transforming technology in biology. In this paper, we study a population genomics problem motivated by the newly available short reads data from high-throughput sequencing. In this problem, we are given short reads collected from individuals in a population. The objective is to infer haplotypes with the given reads. We first formulate the computational problem of haplotype inference with short reads. Based on a simple probabilistic model on short reads, we present a new approach of inferring haplotypes directly from given reads (i.e. without first calling genotypes). Our method is finding the most likely haplotypes whose local genealogical history can be approximately modeled as a perfect phylogeny. We show that the optimal haplotypes under this objective can be found for many data using integer linear programming for modest sized data when there is no recombination. We then develop a related heuristic method which can work with larger data, and also allows recombination. Simulation shows that the performance of our method is competitive against alternative approaches.

The following sections are included:

• Introduction

• Session Papers

• Acknowledgments

• References

This paper describes a scheme for implementing a binary counter with chemical reactions. The value of the counter is encoded by logical values of "0" and "1" that correspond to the absence and presence of specific molecular types, respectively. It is incremented when molecules of a trigger type are injected. Synchronization is achieved with reactions that produce a sustained three-phase oscillation. This oscillation plays a role analogous to a clock signal in digital electronics. Quantities are transferred between molecular types in different phases of the oscillation. Unlike all previous schemes for chemical computation, this scheme is dependent only on coarse rate categories for the reactions ("fast" and "slow"). Given such categories, the computation is exact and independent of the specific reaction rates. Although conceptual for the time being, the methodology has potential applications in domains of synthetic biology such as biochemical sensing and drug delivery. We are exploring DNA-based computation via strand displacement as a possible experimental chassis.

Determining biological network dependencies that can help predict the behavior of a system given prior observations from high-throughput data is a very valuable but difficult task, especially in the light of the ever-increasing volume of experimental data. Such an endeavor can be greatly enhanced by considering regulatory influences on co-expressed groups of genes representing functional modules, thus constraining the number of parameters in the system. This allows development of network models that are predictive of system dynamics. We first develop a predictive network model of the transcriptomics of whole blood from a mouse model of neuroprotection in ischemic stroke, and show that it can accurately predict system behavior under novel conditions. We then use a network topology approach to expand the set of regulators considered and show that addition of topological bottlenecks improves the performance of the predictive model. Finally, we explore how improvements in definition of functional modules may be achieved through an integration of inferred network relationships and functional relationships defined using Gene Ontology similarity. We show that appropriate integration of these two types of relationships can result in models with improved performance.

This paper presents a collection of computational modules implemented with chemical reactions: an inverter, an incrementer, a decrementer, a copier, a comparator, and a multiplier. Unlike previous schemes for chemical computation, ours produces designs that are dependent only on coarse rate categories for the reactions ("fast" vs. "slow"). Given such categories, the computation is exact and independent of the specific reaction rates. We validate our designs through stochastic simulations of the chemical kinetics. Although conceptual for the time being, our methodology has potential applications in domains of synthetic biology such as biochemical sensing and drug delivery. We are exploring DNA-based computation via strand displacement as a possible experimental chassis.

To decipher the dynamical functioning of cellular processes, the method of choice is to observe the time response of cells subjected to well controlled perturbations in time and amplitude. Efficient methods, based on molecular biology, are available to monitor quantitatively and dynamically many cellular processes. In contrast, it is still a challenge to perturb cellular processes - such as gene expression - in a precise and controlled manner. Here, we propose a first step towards in vivo control of gene expression: in real-time, we dynamically control the activity of a yeast signaling cascade thanks to an experimental platform combining a micro-fluidic device, an epi-fluorescence microscope and software implementing control approaches. We experimentally demonstrate the feasibility of this approach, and we investigate computationally some possible improvements of our control strategy using a model of the yeast osmo-adaptation response fitted to our data.

Motivation: A grand challenge in the modeling of biological systems is the identification of key variables which can act as targets for intervention. Good intervention targets are the "key players" in a system and have significant influence over other variables; in other words, in the context of diseases such as cancer, targeting these variables with treatments and interventions will provide the greatest effects because of their direct and indirect control over other parts of the system. Boolean networks are among the simplest of models, yet they have been shown to adequately model many of the complex dynamics of biological systems. Often ignored in the Boolean network model, however, are the so called basins of attraction. As the attractor states alone have been shown to correspond to cellular phenotypes, it is logical to ask which variables are most responsible for triggering a path through a basin to a particular attractor.

Results: This work claims that logic minimization (i.e. classical circuit design) of the collections of states in Boolean network basins of attraction reveals key players in the network. Furthermore, we claim that the key players identified by this method are often excellent targets for intervention given a network modeling a biological system, and more importantly, that the key players identified are not apparent from the attractor states alone, from existing Boolean network measures, or from other network measurements. We demonstrate these claims with a well-studied yeast cell cycle network and with a WNT5A network for melanoma, computationally predicted from gene expression data.

The promise of pharmacogenomics for individualized medicine is on the crest of realization as a result of advances that allow us to predict beneficial, non-beneficial, and deleterious drugs for specific individuals based on aspects of both the individual and the drug. In spite of these advances, information management in this field relies on fairly traditional means, which do not scale to the available volume of full text publications. The aim of this workshop is to bring together researchers working on the automatic or semi-automatic extraction of relationships between biomedical entities from the pharmacogenomic research literature. The workshop will focus particularly on methods for the extraction of genotype-phenotype, genotype-drug, and phenotype-drug relationships and the use of the relationships for advancing pharmacogenomic research. Efforts aimed at creating benchmark corpora as well as comparative evaluation of existing relationship extraction methods are of special interest.

The workshop focused on approaches to deduce changes in biological activity in cellular pathways and networks that drive phenotype from high-throughput data. Work in cancer has demonstrated conclusively that cancer etiology is driven not by single gene mutation or expression change, but by coordinated changes in multiple signaling pathways. These pathway changes involve different genes in different individuals, leading to the failure of gene-focused analysis to identify the full range of mutations or expression changes driving cancer development. There is also evidence that metabolic pathways rather than individual genes play the critical role in a number of metabolic diseases. Tools to look at pathways and networks are needed to improve our understanding of disease and to improve our ability to target therapeutics at appropriate points in these pathways.

Electron cryo-microscopy (cryoEM) is a rapidly maturing methodology in structural biology, which now enables the determination of 3D structures of molecules, macromolecular complexes and cellular components at resolutions as high as 3.5Å, bridging the gap between light microscopy and X-ray crystallography/NMR. In recent years structures of many complex molecular machines have been visualized using this method. Single particle reconstruction, the most widely used technique in cryoEM, has recently demonstrated the capability of producing structures at resolutions approaching those of X-ray crystallography, with over a dozen structures at better than 5 Å resolution published to date. This method represents a significant new source of experimental data for molecular modeling and simulation studies. CryoEM derived maps and models are archived through EMDataBank.org joint deposition services to the EM Data Bank (EMDB) and Protein Data Bank (PDB), respectively. CryoEM maps are now being routinely produced over the 3 - 30 Å resolution range, and a number of computational groups are developing software for building coordinate models based on this data and developing validation techniques to better assess map and model accuracy. In this workshop we will present the results of the first cryoEM modeling challenge, in which computational groups were asked to apply their tools to a selected set of published cryoEM structures. We will also compare the results of the various applied methods, and discuss the current state of the art and how we can most productively move forward.