Biocomputing 2014

The following sections are included:

• CONTENTS

• PREFACE

Precision medicine promises to transform cancer treatment in the next decade through the use of high-throughput sequencing and other technologies to identify telltale molecular aberrations that reveal therapeutic vulnerabilities of each patient's tumor [1]. This session will address the “panomics” of cancer – the complex combination of patient-specific characteristics that drive the development of each person's tumor and response to therapy [2]. The realization of this vision will require novel infrastructure and computational methods to integrate large-scale data effectively and query it in real-time for therapy and/or clinical trial selection for each patient…

The growing availability of inexpensive high-throughput sequence data is enabling researchers to sequence tumor populations within a single individual at high coverage. But, cancer genome sequence evolution and mutational phenomena like driver mutations and gene fusions are difficult to investigate without first reconstructing tumor haplotype sequences. Haplotype assembly of single individual tumor populations is an exceedingly difficult task complicated by tumor haplotype heterogeneity, tumor or normal cell sequence contamination, polyploidy, and complex patterns of variation. While computational and experimental haplotype phasing of diploid genomes has seen much progress in recent years, haplotype assembly in cancer genomes remains uncharted territory.

In this work, we describe HapCompass-Tumor a computational modeling and algorithmic framework for haplotype assembly of copy number variable cancer genomes containing haplotypes at different frequencies and complex variation. We extend our polyploid haplotype assembly model and present novel algorithms for (1) complex variations, including copy number changes, as varying numbers of disjoint paths in an associated graph, (2) variable haplotype frequencies and contamination, and (3) computation of tumor haplotypes using simple cycles of the compass graph which constrain the space of haplotype assembly solutions. The model and algorithm are implemented in the software package HapCompass-Tumor which is available for download from http://www.brown.edu/Research/Istrail_Lab/.

We present a joint analysis method for mutation and gene expression data employing information about proteins that are highly interconnected at the level of protein to protein (pp) interactions, which we apply to the TCGA Acute Myeloid Leukemia (AML) dataset. Given the low incidence of most mutations in virtually all cancer types, as well as the significant inter-patient heterogeneity of the mutation landscape, determining the true causal mutations in each individual patient remains one of the most important challenges for personalized cancer diagnostics and therapy. More automated methods are needed for determining these “driver” mutations in each individual patient. For this purpose, we are exploiting two types of contextual information: (1) the pp interactions of the mutated genes, as well as (2) their potential correlations with gene expression clusters. The use of pp interactions is based on our surprising finding that most AML mutations tend to affect nontrivial protein to protein interaction cliques.

Computational efficiency is important for learning algorithms operating in the “large p, small n” setting. In computational biology, the analysis of data sets containing tens of thousands of features (“large p”), but only a few hundred samples (“small n”), is nowadays routine, and regularized regression approaches such as ridge-regression, lasso, and elastic-net are popular choices. In this paper we propose a novel and highly efficient Bayesian inference method for fitting ridge-regression. Our method is fully analytical, and bypasses the need for expensive tuning parameter optimization, via cross-validation, by employing Bayesian model averaging over the grid of tuning parameters. Additional computational efficiency is achieved by adopting the singular value decomposition reparametrization of the ridge-regression model, replacing computationally expensive inversions of large p × p matrices by efficient inversions of small and diagonal n × n matrices. We show in simulation studies and in the analysis of two large cancer cell line data panels that our algorithm achieves slightly better predictive performance than cross-validated ridge-regression while requiring only a fraction of the computation time. Furthermore, in comparisons based on the cell line data sets, our algorithm systematically out-performs the lasso in both predictive performance and computation time, and shows equivalent predictive performance, but considerably smaller computation time, than the elastic-net.

Advances in experimental techniques resulted in abundant genomic, transcriptomic, epigenomic, and proteomic data that have the potential to reveal critical drivers of human diseases. Complementary algorithmic developments enable researchers to map these data onto protein-protein interaction networks and infer which signaling pathways are perturbed by a disease. Despite this progress, integrating data across different biological samples or patients remains a substantial challenge because samples from the same disease can be extremely heterogeneous. Somatic mutations in cancer are an infamous example of this heterogeneity. Although the same signaling pathways may be disrupted in a cancer patient cohort, the distribution of mutations is long-tailed, and many driver mutations may only be detected in a small fraction of patients. We developed a computational approach to account for heterogeneous data when inferring signaling pathways by sharing information across the samples. Our technique builds upon the prize-collecting Steiner forest problem, a network optimization algorithm that extracts pathways from a protein-protein interaction network. We recover signaling pathways that are similar across all samples yet still reflect the unique characteristics of each biological sample. Leveraging data from related tumors improves our ability to recover the disrupted pathways and reveals patient-specific pathway perturbations in breast cancer.

The two-hit model of carcinogenesis provides a valuable framework for understanding the role of DNA repair and tumor suppressor genes in cancer development and progression. Under this model, tumor development can initiate from a single somatic mutation in individuals that inherit an inactivating germline variant. Although the two-hit model can be an overgeneralization, the tendency for the pattern of somatic mutations to differ in cancer patients that inherit predisposition alleles is a signal that can be used to identify and validate germline susceptibility variants. Here, we present the Somatic-Germline Interaction (SGI) tool, which is designed to identify statistical interaction between germline variants and somatic mutational events from next-generation sequence data. SGI interfaces with rare-variant association tests and variant classifiers to identify candidate germline susceptibility variants from case-control sequencing data. SGI then analyzes tumor-normal pair next-generation sequence data to evaluate evidence for somatic-germline interaction in each gene or pathway using two tests: the Allelic Imbalance Rank Sum (AIRS) test and the Somatic Mutation Interaction Test (SMIT). AIRS tests for preferential allelic imbalance to evaluate whether somatic mutational events tend to amplify candidate germline variants. SMIT evaluates whether somatic point mutations and small indels occur more or less frequently than expected in the presence of candidate germline variants. Both AIRS and SMIT control for heterogeneity in the mutational process resulting from regional variation in mutation rates and inter-sample variation in background mutation rates. The SGI test combines AIRS and SMIT to provide a single, unified measure of statistical interaction between somatic mutational events and germline variation. We show that the tests implemented in SGI have high power with relatively modest sample sizes in a wide variety of scenarios. We demonstrate the utility of SGI to increase the power of rare variant association studies in cancer and to validate the potential role in cancer causation of germline susceptibility variants.

Large-scale pharmacogenomic screens of cancer cell lines have emerged as an attractive pre-clinical system for identifying tumor genetic subtypes with selective sensitivity to targeted therapeutic strategies. Application of modern machine learning approaches to pharmacogenomic datasets have demonstrated the ability to infer genomic predictors of compound sensitivity. Such modeling approaches entail many analytical design choices; however, a systematic study evaluating the relative performance attributable to each design choice is not yet available. In this work, we evaluated over 110,000 different models, based on a multifactorial experimental design testing systematic combinations of modeling factors within several categories of modeling choices, including: type of algorithm, type of molecular feature data, compound being predicted, method of summarizing compound sensitivity values, and whether predictions are based on discretized or continuous response values. Our results suggest that model input data (type of molecular features and choice of compound) are the primary factors explaining model performance, followed by choice of algorithm. Our results also provide a statistically principled set of recommended modeling guidelines, including: using elastic net or ridge regression with input features from all genomic profiling platforms, most importantly, gene expression features, to predict continuous-valued sensitivity scores summarized using the area under the dose response curve, with pathway targeted compounds most likely to yield the most accurate predictors. In addition, our study provides a publicly available resource of all modeling results, an open source code base, and experimental design for researchers throughout the community to build on our results and assess novel methodologies or applications in related predictive modeling problems.

Cancer cells derived from different stages of tumor progression may exhibit distinct biological properties, as exemplified by the paired lung cancer cell lines H1993 and H2073. While H1993 was derived from chemo-naive metastasized tumor, H2073 originated from the chemo-resistant primary tumor from the same patient and exhibits strikingly different drug response profile. To understand the underlying genetic and epigenetic bases for their biological properties, we investigated these cells using a wide range of large-scale methods including whole genome sequencing, RNA sequencing, SNP array, DNA methylation array, and de novo genome assembly. We conducted an integrative analysis of both cell lines to distinguish between potential driver and passenger alterations. Although many genes are mutated in these cell lines, the combination of DNA- and RNA-based variant information strongly implicates a small number of genes including TP53 and STK11 as likely drivers. Likewise, we found a diverse set of genes differentially expressed between these cell lines, but only a fraction can be attributed to changes in DNA copy number or methylation. This set included the ABC transporter ABCC4, implicated in drug resistance, and the metastasis associated MET oncogene. While the rich data content allowed us to reduce the space of hypotheses that could explain most of the observed biological properties, we also caution there is a lack of statistical power and inherent limitations in such single patient case studies.

Disrupted or abnormal biological processes responsible for cancers often quantitatively manifest as disrupted additive and multiplicative interactions of gene/protein expressions correlating with cancer progression. However, the examination of all possible combinatorial interactions between gene features in most case-control studies with limited training data is computationally infeasible. In this paper, we propose a practically feasible data integration approach, QUIRE (QUadratic Interactions among infoRmative fEatures), to identify discriminative complex interactions among informative gene features for cancer diagnosis and biomarker discovery directly based on patient blood samples. QUIRE works in two stages, where it first identifies functionally relevant gene groups for the disease with the help of gene functional annotations and available physical protein interactions, then it explores the combinatorial relationships among the genes from the selected informative groups. Based on our private experimentally generated data from patient blood samples using a novel SOMAmer (Slow Off-rate Modified Aptamer) technology, we apply QUIRE to cancer diagnosis and biomarker discovery for Renal Cell Carcinoma (RCC) and Ovarian Cancer (OVC). To further demonstrate the general applicability of our approach, we also apply QUIRE to a publicly available Colorectal Cancer (CRC) dataset that can be used to prioritize our SOMAmer design. Our experimental results show that QUIRE identifies gene-gene interactions that can better identify the different cancer stages of samples, as compared to other state-of-the-art feature selection methods. A literature survey shows that many of the interactions identified by QUIRE play important roles in the development of cancer.

We propose and discuss a method for doing gene expression meta-analysis (multiple datasets) across multiplex measurement modalities measuring the expression of many genes simultaneously (e.g. microarrays and RNAseq) using external control samples and a method of heterogeneity detection to identify and filter on comparable gene expression measurements. We demonstrate this approach on publicly available gene expression datasets from samples of medulloblastoma and normal cerebellar tissue and identify some potential new targets in the treatment of medulloblastoma.

Despite increasing investments in pharmaceutical R&D, there is a continuing paucity of new drug approvals. Drug discovery continues to be a lengthy and resource-consuming process in spite of all the advances in genomics, life sciences, and technology. Indeed, it is estimated that about 90% of the drugs fail during development in phase 1 clinical trials¹ and that it takes billions of dollars in investment and an average of 15 years to bring a new drug to the market²…

Repurposing an existing drug for an alternative use is not only a cost effective method of development, but also a faster process due to the drug's previous clinical testing and established pharmokinetic profiles. A potentially rich resource for computational drug repositioning approaches is publically available high throughput screening data, available in databases such as PubChem Bioassay and ChemBank. We examine statistical and computational considerations for secondary analysis of publicly available high throughput screening (HTS) data with respect to metadata, data quality, and completeness. We discuss developing methods and best practices that can help to ameliorate these issues.

The revolution in sequencing techniques in the past decade has provided an extensive picture of the molecular mechanisms behind complex diseases such as cancer. The Cancer Cell Line Encyclopedia (CCLE) and The Cancer Genome Project (CGP) have provided an unprecedented opportunity to examine copy number, gene expression, and mutational information for over 1000 cell lines of multiple tumor types alongside IC50 values for over 150 different drugs and drug related compounds. We present a novel pipeline called DIRPP, Drug Intervention Response Predictions with PARADIGM⁷, which predicts a cell line's response to a drug intervention from molecular data. PARADIGM (Pathway Recognition Algorithm using Data Integration on Genomic Models) is a probabilistic graphical model used to infer patient specific genetic activity by integrating copy number and gene expression data into a factor graph model of a cellular network. We evaluated the performance of DIRPP on endometrial, ovarian, and breast cancer related cell lines from the CCLE and CGP for nine drugs. The pipeline is sensitive enough to predict the response of a cell line with accuracy and precision across datasets as high as 80 and 88% respectively. We then classify drugs by the specific pathway mechanisms governing drug response. This classification allows us to compare drugs by cellular response mechanisms rather than simply by their specific gene targets. This pipeline represents a novel approach for predicting clinical drug response and generating novel candidates for drug repurposing and repositioning.

The emergence of multi-drug and extensive drug resistance of microbes to antibiotics poses a great threat to human health. Although drug repurposing is a promising solution for accelerating the drug development process, its application to anti-infectious drug discovery is limited by the scope of existing phenotype-, ligand-, or target-based methods. In this paper we introduce a new computational strategy to determine the genome-wide molecular targets of bioactive compounds in both human and bacterial genomes. Our method is based on the use of a novel algorithm, ligand Enrichment of Network Topological Similarity (ligENTS), to map the chemical universe to its global pharmacological space. ligENTS outperforms the state-of-the-art algorithms in identifying novel drug-target relationships. Furthermore, we integrate ligENTS with our structural systems biology platform to identify drug repurposing opportunities via target similarity profiling. Using this integrated strategy, we have identified novel P. falciparum targets of drug-like active compounds from the Malaria Box, and suggest that a number of approved drugs may be active against malaria. This study demonstrates the potential of an integrative chemical genomics and structural systems biology approach to drug repurposing.

In silico prediction of unknown drug-target interactions (DTIs) has become a popular tool for drug repositioning and drug development. A key challenge in DTI prediction lies in integrating multiple types of data for accurate DTI prediction. Although recent studies have demonstrated that genomic, chemical and pharmacological data can provide reliable information for DTI prediction, it remains unclear whether functional information on proteins can also contribute to this task. Little work has been developed to combine such information with other data to identify new interactions between drugs and targets. In this paper, we introduce functional data into DTI prediction and construct biological space for targets using the functional similarity measure. We present a probabilistic graphical model, called conditional random field (CRF), to systematically integrate genomic, chemical, functional and pharmacological data plus the topology of DTI networks into a unified framework to predict missing DTIs. Tests on two benchmark datasets show that our method can achieve excellent prediction performance with the area under the precision-recall curve (AUPR) up to 94.9. These results demonstrate that our CRF model can successfully exploit heterogeneous data to capture the latent correlations of DTIs, and thus will be practically useful for drug repositioning. Supplementary Material is available at http://iiis.tsinghua.edu.cn/~compbio/papers/psb2014/psb2014_sm.pdf.

We present a probabilistic data fusion framework that combines multiple computational approaches for drawing relationships between drugs and targets. The approach has special relevance to identifying surprising unintended biological targets of drugs. Comparisons between molecules are made based on 2D topological structural considerations, based on 3D surface characteristics, and based on English descriptions of clinical effects. Similarity computations within each modality were transformed into probability scores. Given a new molecule along with a set of molecules sharing some biological effect, a single score based on comparison to the known set is produced, reflecting either 2D similarity, 3D similarity, clinical effects similarity or their combination. The methods were validated within acurated structural pharmacology database (SPDB) and further tested by blind application to data derived from the ChEMBL database. For prediction of off-target effects, 3D-similarity performed best as a single modality, but combining all methods produced performance gains. Striking examples of structurally surprising off-target predictions are presented.

Computational drug repositioning leverages computational technology and high volume of biomedical data to identify new indications for existing drugs. Since it does not require costly experiments that have a high risk of failure, it has attracted increasing interest from diverse fields such as biomedical, pharmaceutical, and informatics areas. In this study, we used pharmacogenomics data generated from pharmacogenomics studies, applied informatics and Semantic Web technologies to address the drug repositioning problem. Specifically, we explored PharmGKB to identify pharmacogenomics related associations as pharmacogenomics profiles for US Food and Drug Administration (FDA) approved breast cancer drugs. We then converted and represented these profiles in Semantic Web notations, which support automated semantic inference. We successfully evaluated the performance and efficacy of the breast cancer drug pharmacogenomics profiles by case studies. Our results demonstrate that combination of pharmacogenomics data and Semantic Web technology/Cheminformatics approaches yields better performance of new indication and possible adverse effects prediction for breast cancer drugs.

The following sections are included:

• Introduction
  • Defining Genetic Elements
  • Defining Phenotypes
  • The Prevalence of Pleiotropy
  • Pleiotropy in an Evolutionary Context
  • Pleiotropy and Human Health

• Session Contributions

• Acknowledgments

• References

With the rapid increase in the quality and quantity of data generated by modern high-throughput sequencing techniques, there has been a need for innovative methods able to convert this tremendous amount of data into more accessible forms. Networks have been a corner stone of this movement, as they are an intuitive way of representing interaction data, yet they offer a full set of sophisticated statistical tools to analyze the phenomena they model. We propose a novel approach to reveal and analyze pleiotropic and epistatic effects at the genome-wide scale using a bipartite network composed of human diseases, phenotypic traits, and several types of predictive elements (i.e. SNPs, genes, or pathways). We take advantage of publicly available GWAS data, gene and pathway databases, and more to construct networks different levels of granularity, from common genetic variants to entire biological pathways. We use the connections between the layers of the network to approximate the pleiotropy and epistasis effects taking place between the traits and the predictive elements. The global graph-theory based quantitative methods reveal that the levels of pleiotropy and epistasis are comparable for all types of predictive element. The results of the magnified “glaucoma” region of the network demonstrate the existence of well documented interactions, supported by overlapping genes and biological pathway, and more obscure associations. As the amount and complexity of genetic data increases, bipartite, and more generally multipartite networks that combine human diseases and other physical attributes with layers of genetic information, have the potential to become ubiquitous tools in the study of complex genetic and phenotypic interactions.

Environment-wide association studies (EWAS) provide a way to uncover the environmental mechanisms involved in complex traits in a high-throughput manner. Genome-wide association studies have led to the discovery of genetic variants associated with many common diseases but do not take into account the environmental component of complex phenotypes. This EWAS assesses the comprehensive association between environmental variables and the outcome of type 2 diabetes (T2D) in the Marshfield Personalized Medicine Research Project Biobank (Marshfield PMRP). We sought replication in two National Health and Nutrition Examination Surveys (NHANES). The Marshfield PMRP currently uses four tools for measuring environmental exposures and outcome traits: 1) the PhenX Toolkit includes standardized exposure and phenotypic measures across several domains, 2) the Diet History Questionnaire (DHQ) is a food frequency questionnaire, 3) the Measurement of a Person's Habitual Physical Activity scores the level of an individual's physical activity, and 4) electronic health records (EHR) employs validated algorithms to establish T2D case-control status. Using PLATO software, 314 environmental variables were tested for association with T2D using logistic regression, adjusting for sex, age, and BMI in over 2,200 European Americans. When available, similar variables were tested with the same methods and adjustment in samples from NHANES III and NHANES 1999-2002. Twelve and 31 associations were identified in the Marshfield samples at p<0.01 and p<0.05, respectively. Seven and 13 measures replicated in at least one of the NHANES at p<0.01 and p<0.05, respectively, with the same direction of effect. The most significant environmental exposures associated with T2D status included decreased alcohol use as well as increased smoking exposure in childhood and adulthood. The results demonstrate the utility of the EWAS method and survey tools for identifying environmental components of complex diseases like type 2 diabetes. These high-throughput and comprehensive investigation methods can easily be applied to investigate the relation between environmental exposures and multiple phenotypes in future analyses.

Global transcript expression experiments are commonly used to investigate the biological processes that underlie complex traits. These studies can exhibit complex patterns of pleiotropy when trans-acting genetic factors influence overlapping sets of multiple transcripts. Dissecting these patterns into biological modules with distinct genetic etiology can provide models of how genetic variants affect specific processes that contribute to a trait. Here we identify transcript modules associated with pleiotropic genetic factors and apply genetic interaction analysis to disentangle the regulatory architecture in a mouse intercross study of kidney function. The method, called the combined analysis of pleiotropy and epistasis (CAPE), has been previously used to model genetic networks for multiple physiological traits. It simultaneously models multiple phenotypes to identify direct genetic influences as well as influences mediated through genetic interactions. We first identify candidate trans expression quantitative trait loci (eQTL) and the transcripts potentially affected. We then clustered the transcripts into modules of co-expressed genes, from which we compute summary module phenotypes. Finally, we applied CAPE to map the network of interacting module QTL (modQTL) affecting the gene modules. The resulting network mapped how multiple modQTL both directly and indirectly affect modules associated with metabolic functions and biosynthetic processes. This work demonstrates how the integration of pleiotropic signals in gene expression data can be used to infer a complex hypothesis of how multiple loci interact to co-regulate transcription programs, thereby providing additional constraints to prioritize validation experiments.

The following sections are included:

• Session contributions

• References

The American College of Medical Genetics and Genomics (ACMG) recently released guidelines regarding the reporting of incidental findings in sequencing data. Given the availability of Direct to Consumer (DTC) genetic testing and the falling cost of whole exome and genome sequencing, individuals will increasingly have the opportunity to analyze their own genomic data. We have developed a web-based tool, PATH-SCAN, which annotates individual genomes and exomes for ClinVar designated pathogenic variants found within the genes from the ACMG guidelines. Because mutations in these genes predispose individuals to conditions with actionable outcomes, our tool will allow individuals or researchers to identify potential risk variants in order to consult physicians or genetic counselors for further evaluation. Moreover, our tool allows individuals to anonymously submit their pathogenic burden, so that we can crowd source the collection of quantitative information regarding the frequency of these variants. We tested our tool on 1092 publicly available genomes from the 1000 Genomes project, 163 genomes from the Personal Genome Project, and 15 genomes from a clinical genome sequencing research project. Excluding the most commonly seen variant in 1000 Genomes, about 20% of all genomes analyzed had a ClinVar designated pathogenic variant that required further evaluation.

A striking finding from recent large-scale sequencing efforts is that the vast majority of variants in the human genome are rare and found within single populations or lineages. These observations hold important implications for the design of the next round of disease variant discovery efforts—if genetic variants that influence disease risk follow the same trend, then we expect to see population-specific disease associations that require large sample sizes for detection. To address this challenge, and due to the still prohibitive cost of sequencing large cohorts, researchers have developed a new generation of low-cost genotyping arrays that assay rare variation previously identified from large exome sequencing studies. Genotyping approaches rely not only on directly observing variants, but also on phasing and imputation methods that use publicly available reference panels to infer unobserved variants in a study cohort. Rare variant exome arrays are intentionally enriched for variants likely to be disease causing, and here we assay the ability of the first commercially available rare exome variant array (the Illumina Infinium HumanExome BeadChip) to also tag other potentially damaging variants not molecularly assayed. Using full sequence data from chromosome 22 from the phase I 1000 Genomes Project, we evaluate three methods for imputation (BEAGLE, MaCH-Admix, and SHAPEIT2/IMPUTE2) with the rare exome variant array under varied study panel sizes, reference panel sizes, and LD structures via population differences. We find that imputation is more accurate across both the genome and exome for common variant arrays than the next generation array for all allele frequencies, including rare alleles. We also find that imputation is the least accurate in African populations, and accuracy is substantially improved for rare variants when the same population is included in the reference panel. Depending on the goals of GWAS researchers, our results will aid budget decisions by helping determine whether money is best spent sequencing the genomes of smaller sample sizes, genotyping larger sample sizes with rare and/or common variant arrays and imputing SNPs, or some combination of the two.

Calcineurin-inhibitors CI are immunosuppressive agents prescribed to patients after solid organ transplant to prevent rejection. Although these drugs have been transformative for allograft survival, long-term use is complicated by side effects including nephrotoxicity. Given the narrow therapeutic index of CI, therapeutic drug monitoring is used to prevent acute rejection from underdosing and acute toxicity from overdosing, but drug monitoring does not alleviate long-term side effects. Patients on calcineurin-inhibitors for long periods almost universally experience declines in renal function, and a subpopulation of transplant recipients ultimately develop chronic kidney disease that may progress to end stage renal disease attributable to calcineurin inhibitor toxicity (CNIT). Pharmacogenomics has the potential to identify patients who are at high risk for developing advanced chronic kidney disease caused by CNIT and providing them with existing alternate immunosuppressive therapy. In this study we utilized BioVU, Vanderbilt University Medical Center's DNA biorepository linked to de-identified electronic medical records to identify a cohort of 115 heart transplant recipients prescribed calcineurin-inhibitors to identify genetic risk factors for CNIT We identified 37 cases of nephrotoxicity in our cohort, defining nephrotoxicity as a monthly median estimated glomerular filtration rate (eGFR) <30 mL/min/1.73m2 at least six months post-transplant for at least three consecutive months. All heart transplant patients were genotyped on the Illumina ADME Core Panel, a pharmacogenomic genotyping platform that assays 184 variants across 34 genes. In Cox regression analysis adjusting for age at transplant, pre-transplant chronic kidney disease, pre-transplant diabetes, and the three most significant principal components (PCAs), we did not identify any markers that met our multiple-testing threshold. As a secondary analysis we also modeled post-transplant eGFR directly with linear mixed models adjusted for age at transplant, cyclosporine use, median BMI, and the three most significant principal components. While no SNPs met our threshold for significance, a SNP previously identified in genetic studies of the dosing of tacrolimus CYP34A rs776746, replicated in an adjusted analysis at an uncorrected p-value of 0.02 (coeff(S.E.) = 14.60(6.41)). While larger independent studies will be required to further validate this finding, this study underscores the EMRs usefulness as a resource for longitudinal pharmacogenetic study designs.

Simultaneously reverse engineering a collection of condition-specific gene networks from gene expression microarray data to uncover dynamic mechanisms is a key challenge in systems biology. However, existing methods for this task are very sensitive to variations in the size of the microarray samples across different biological conditions (which we term sample size heterogeneity in network reconstruction), and can potentially produce misleading results that can lead to incorrect biological interpretation. In this work, we develop a more robust framework that addresses this novel problem. Just like microarray measurements across conditions must undergo proper normalization on their magnitudes before entering subsequent analysis, we argue that networks across conditions also need to be “normalized” on their density when they are constructed, and we provide an algorithm that allows such normalization to be facilitated while estimating the networks. We show the quantitative advantages of our approach on synthetic and real data. Our analysis of a hematopoietic stem cell dataset reveals interesting results, some of which are confirmed by previously validated results.

In case-control studies of rare Mendelian disorders and complex diseases, the power to detect variant and gene-level associations of a given effect size is limited by the size of the study sample. Paradoxically, low statistical power may increase the likelihood that a statistically significant finding is also a false positive. The prioritization of variants based on call quality, putative effects on protein function, the predicted degree of deleteriousness, and allele frequency is often used as a mechanism for reducing the occurrence of false positives, while preserving the set of variants most likely to contain true disease associations. We propose that specificity can be further improved by considering errors that are specific to the regions of the genome being sequenced. These problematic regions (PRs) are identified a-priori and are used to down-weight constitutive variants in a case-control analysis. Using samples drawn from 1000-Genomes, we illustrate the utility of PRs in identifying true variant and gene associations using a case-control study on a known Mendelian disease, cystic fibrosis(CF).

The immune system gathers evidence of the execution of various molecular processes, both foreign and the cells' own, as time- and space-varying sets of epitopes, small linear or conformational segments of the proteins involved in these processes. Epitopes do not have any obvious ordering in this scheme: The immune system simply sees these epitope sets as disordered “bags” of simple signatures based on whose contents the actions need to be decided. The immense landscape of possible bags of epitopes is shaped by the cellular pathways in various cells, as well as the characteristics of the internal sampling process that chooses and brings epitopes to cellular surface. As a consequence, upon the infection by the same pathogen, different individuals' cells present very different epitope sets. Modeling this landscape should thus be a key step in computational immunology. We show that among possible bag-of-words models, the counting grid is most fit for modeling cellular presentation. We describe each patient by a bag-of-peptides they are likely to present on the cellular surface. In regression tests, we found that compared to the state-of-the-art, counting grids explain more than twice as much of the log viral load variance in these patients. This is potentially a significant advancement in the field, given that a large part of the log viral load variance also depends on the infecting HIV strain, and that HIV polymorphisms themselves are known to strongly associate with HLA types, both effects beyond what is modeled here.

A key step for Alzheimer's disease (AD) study is to identify associations between genetic variations and intermediate phenotypes (e.g., brain structures). At the same time, it is crucial to develop a noninvasive means for AD diagnosis. Although these two tasks—association discovery and disease diagnosis—have been treated separately by a variety of approaches, they are tightly coupled due to their common biological basis. We hypothesize that the two tasks can potentially benefit each other by a joint analysis, because (i) the association study discovers correlated biomarkers from different data sources, which may help improve diagnosis accuracy, and (ii) the disease status may help identify disease-sensitive associations between genetic variations and MRI features. Based on this hypothesis, we present a new sparse Bayesian approach for joint association study and disease diagnosis. In this approach, common latent features are extracted from different data sources based on sparse projection matrices and used to predict multiple disease severity levels based on Gaussian process ordinal regression; in return, the disease status is used to guide the discovery of relationships between the data sources. The sparse projection matrices not only reveal the associations but also select groups of biomarkers related to AD. To learn the model from data, we develop an efficient variational expectation maximization algorithm. Simulation results demonstrate that our approach achieves higher accuracy in both predicting ordinal labels and discovering associations between data sources than alternative methods. We apply our approach to an imaging genetics dataset of AD. Our joint analysis approach not only identifies meaningful and interesting associations between genetic variations, brain structures, and AD status, but also achieves significantly higher accuracy for predicting ordinal AD stages than the competing methods.

Text and data mining methods constantly advance and are applied in different fields. In order for them to impact the biomedical discovery process, it is necessary to thoroughly engage scientists at both ends, and conduct thorough empirical evaluations as to their ability to suggest novel hypotheses and address the most crucial questions. The PSB 2014 Session on Text and Data Mining for Biomedical Discovery presents eight papers that advance the field in this mutually reinforcing fashion. Work presented in this session includes data mining and analysis techniques that are applicable to abroad spectrum of problems, including the analysis and visualization of mass spectrometry based proteomics data and longitudinal data, as well as gene function, protein function and protein fold prediction. Text mining approaches selected for presentation include a method for predicting genes involved in disease or in drug response, a method for extracting events relevant to biological pathways, and an approach that mixes text and data mining techniques to predict important milestones in the female reproductive lifespan.

We propose a new kernel-based method for the classification of protein sequences and structures. We first represent each protein as a set of time series data using several structural, physicochemical, and predicted properties such as a sequence of consecutive dihedral angles, hydrophobicity indices, or predictions of disordered regions. A kernel function is then computed for pairs of proteins, exploiting the principles of vector quantization and subsequently used with support vector machines for protein classification. Although our method requires a significant pre-processing step, it is fast in the training and prediction stages owing to the linear complexity of kernel computation with the length of protein sequences. We evaluate our approach on two protein classification tasks involving the prediction of SCOP structural classes and catalytic activity according to the Gene Ontology. We provide evidence that the method is competitive when compared to string kernels, and useful for a range of protein classification tasks. Furthermore, the applicability of our approach extends beyond computational biology to any classification of time series data.

Identifying genetic variants that affect drug response or play a role in disease is an important task for clinicians and researchers. Before individual variants can be explored efficiently for effect on drug response or disease relationships, specific candidate genes must be identified. While many methods rank candidate genes through the use of sequence features and network topology, only a few exploit the information contained in the biomedical literature. In this work, we train and test a classifier on known pharmacogenes from PharmGKB and present a classifier that predicts pharmacogenes on a genome-wide scale using only Gene Ontology annotations and simple features mined from the biomedical literature. Performance of F=0.86, AUC=0.860 is achieved. The top 10 predicted genes are analyzed. Additionally, a set of enriched pharmacogenic Gene Ontology concepts is produced.

Mass spectrometry based proteomics technologies have allowed for a great progress in identifying disease biomarkers for clinical diagnosis and prognosis. However, they face acute challenges from a data reproducibility standpoint, in that no two independent studies have been found to produce the same proteomic patterns. Such reproducibility issues cause the identified biomarker patterns to lose repeatability and prevent real clinical usage. In this work, we propose a profile biomarker approach to overcome this problem from a machine-learning viewpoint by developing a novel derivative component analysis (DCA). As an implicit feature selection algorithm, derivative component analysis enables the separation of true signals from red herrings by capturing subtle data behaviors and removing system noises from a proteomic profile. We further demonstrate its advantages in disease diagnosis by viewing input data as a profile biomarker. The results from our profile biomarker diagnosis suggest an effective solution to overcoming proteomics data's reproducibility problem, present an alternative method for biomarker discovery in proteomics, and provide a good candidate for clinical proteomic diagnosis.

The creation of biological pathway knowledge bases is largely driven by manual effort to curate based on evidences from the scientific literature. It is highly challenging for the curators to keep up with the literature. Text mining applications have been developed in the last decade to assist human curators to speed up the curation pace where majority of them aim to identify the most relevant papers for curation with little attempt to directly extract the pathway information from text. In this paper, we describe a rule-based literature mining system to extract pathway information from text. We evaluated the system using curated pharmacokinetic (PK) and pharmacodynamic (PD) pathways in PharmGKB. The system achieved an F-measure of 63.11% and 34.99% for entity extraction and event extraction respectively against all PubMed abstracts cited in PharmGKB. It may be possible to improve the system performance by incorporating using statistical machine learning approaches. This study also helped us gain insights into the barriers towards automated event extraction from text for pathway curation.

Complex diseases such as major depression affect people over time in complicated patterns. Longitudinal data analysis is thus crucial for understanding and prognosis of such diseases and has received considerable attention in the biomedical research community. Traditional classification and regression methods have been commonly applied in a simple (controlled) clinical setting with a small number of time points. However, these methods cannot be easily extended to the more general setting for longitudinal analysis, as they are not inherently built for time-dependent data. Functional regression, in contrast, is capable of identifying the relationship between features and outcomes along with time information by assuming features and/or outcomes as random functions over time rather than independent random variables. In this paper, we propose a novel sparse generalized functional linear model for the prediction of treatment remission status of the depression participants with longitudinal features. Compared to traditional functional regression models, our model enables high-dimensional learning, smoothness of functional coefficients, longitudinal feature selection and interpretable estimation of functional coefficients. Extensive experiments have been conducted on the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) data set and the results show that the proposed sparse functional regression method achieves significantly higher prediction power than existing approaches.

Electronic medical records (EMRs) are becoming more widely implemented following directives from the federal government and incentives for supplemental reimbursements for Medicare and Medicaid claims. Replete with rich phenotypic data, EMRs offer a unique opportunity for clinicians and researchers to identify potential research cohorts and perform epidemiologic studies. Notable limitations to the traditional epidemiologic study include cost, time to complete the study, and limited ancestral diversity; EMR-based epidemiologic studies offer an alternative. The Epidemiologic Architecture for Genes Linked to Environment (EAGLE) Study, as part of the Population Architecture using Genomics and Epidemiology (PAGE) I Study, has genotyped more than 15,000 patients of diverse ancestry in BioVU, the Vanderbilt University Medical Center's biorepository linked to the EMR (EAGLE BioVU). We report here the development and performance of data-mining techniques used to identify the age at menarche (AM) and age at menopause (AAM), important milestones in the reproductive lifespan, in women from EAGLE BioVU for genetic association studies. In addition, we demonstrate the ability to discriminate age at naturally-occurring menopause (ANM) from medically-induced menopause. Unusual timing of these events may indicate underlying pathologies and increased risk for some complex diseases and cancer; however, they are not consistently recorded in the EMR. Our algorithm offers a mechanism by which to extract these data for clinical and research goals.

Label propagation methods are extremely well-suited for a variety of biomedical prediction tasks based on network data. However, these algorithms cannot be used to integrate feature-based data sources with networks. We propose an efficient learning algorithm to integrate these two types of heterogeneous data sources to perform binary prediction tasks on node features (e.g., gene prioritization, disease gene prediction). Our method, LMGraph, consists of two steps. In the first step, we extract a small set of “network features” from the nodes of networks that represent connectivity with labeled nodes in the prediction tasks. In the second step, we apply a simple weighting scheme in conjunction with linear classifiers to combine these network features with other feature data. This two-step procedure allows us to (i) learn highly scalable and computationally efficient linear classifiers, (ii) and seamlessly combine feature-based data sources with networks. Our method is much faster than label propagation which is already known to be computationally efficient on large-scale prediction problems. Experiments on multiple functional interaction networks from three species (mouse, y, C.elegans) with tens of thousands of nodes and hundreds of binary prediction tasks demonstrate the efficacy of our method.

The development of effective methods for the characterization of gene functions that are able to combine diverse data sources in a sound and easily-extendible way is an important goal in computational biology. We have previously developed a general matrix factorization-based data fusion approach for gene function prediction. In this manuscript, we show that this data fusion approach can be applied to gene function prediction and that it can fuse various heterogeneous data sources, such as gene expression profiles, known protein annotations, interaction and literature data. The fusion is achieved by simultaneous matrix tri-factorization that shares matrix factors between sources. We demonstrate the effectiveness of the approach by evaluating its performance on predicting ontological annotations in slime mold D. discoideum and on recognizing proteins of baker's yeast S. cerevisiae that participate in the ribosome or are located in the cell membrane. Our approach achieves predictive performance comparable to that of the state-of-the-art kernel-based data fusion, but requires fewer data preprocessing steps.

The human genome encodes a large number of non-coding RNAs, which employ a new and crucial layer of biological regulation in addition to proteins. Technical advancement in recent years, particularly, the wide application of next generation sequencing analysis, provide an unprecedented opportunity to identify new non-coding RNAs and investigate their functions and regulatory mechanisms. The aim of this workshop is to bring together experimental and computational biologist to exchange ideas on non-coding RNA studies.

Many colleges and universities across the globe now offer bachelors, masters, and doctoral degrees, along with certificate programs in bioinformatics. While there is some consensus surrounding curricula competencies, programs vary greatly in their core foci, with some leaning heavily toward the biological sciences and others toward quantitative areas. This allows prospective students to choose a program that best fits their interests and career goals. In the digital age, most scientific fields are facing an enormous growth of data, and as a consequence, the goals and challenges of bioinformatics are rapidly changing; this requires that bioinformatics education also change. In this workshop, we seek to ascertain current trends in bioinformatics education by asking the question, “What are the core competencies all bioinformaticians should have at the end of their training, and how successful have programs been in placing students in desired careers?”

A clear and predictive understanding of the etiology of autism spectrum disorders (ASD), a group of neurodevelopmental disorders characterized by varying deficits in social interaction and communication as well as repetitive behaviors, has not yet been achieved. There remains active debate about the origins of autism, and the degree to which genetic and environmental factors, and their interplay, produce the range and heterogeneity of cognitive, developmental, and behavioral features seen in children carrying a diagnosis of ASD. Unlocking the causes of these complex developmental disorders will require a collaboration of experts in many disciplines, including clinicians, environmental exposure experts, bioinformaticists, geneticists, and computer scientists. For this workshop we invited prominent researchers in the field of autism, covering a range of topics from genetic and environmental research to ethical considerations. The goal of this workshop: provide an introduction to the current state of autism research, highlighting the potential for multi-disciplinary collaborations that rigorously evaluate the many potential contributors to ASD. It is further anticipated that approaches that successfully advance the understanding of ASD can be applied to the study of other common, complex disorders. Herein we provide a short review of ASD and the work of the invited speakers.