Biocomputing 2015

The following sections are included:

• Preface

PSB brings together top researchers from around the world to exchange research results and address open issues in all aspects of computational biology. PSB 2015 marks the twentieth anniversary of PSB. Reaching a milestone year is an accomplishment well worth celebrating. It is long enough to have seen big changes occur, but recent enough to be relevant for today. As PSB celebrates twenty years of service, we would like to take this opportunity to congratulate the PSB community for your success. We would also like the community to join us in a time of celebration and reflection on this accomplishment.

Targeted cancer treatment is becoming the goal of newly developed oncology medicines and has already shown promise in some spectacular cases such as the case of BRAF kinase inhibitors in BRAF-mutant (e.g. V600E) melanoma. These developments are driven by the advent of high-throughput sequencing, which continues to drop in cost, and that has enabled the sequencing of the genome, transcriptome, and epigenome of the tumors of a large number of cancer patients in order to discover the molecular aberrations that drive the oncogenesis of several types of cancer. Applying these technologies in the clinic promises to transform cancer treatment by identifying therapeutic vulnerabilities of each patient’s tumor. These approaches will need to address the panomics of cancer – the integration of the complex combination of patient-specific characteristics that drive the development of each person’s tumor and response to therapy. This in turn necessitates new computational methods to integrate large-scale “omics” data for each patient with their electronic medical records, and in the context of the results from large-scale pan-cancer research studies, to select the best therapy and/or clinical trial for the patient at hand…

The Cell Index Database, (CELLX) (http://cellx.sourceforge.net) provides a computational framework for integrating expression, copy number variation, mutation, compound activity, and meta data from cancer cells. CELLX provides the computational biologist a quick way to perform routine analyses as well as the means to rapidly integrate data for offline analysis. Data is accessible through a web interface which utilizes R to generate plots and perform clustering, correlations, and statistical tests for associations within and between data types for ~20,000 samples from TCGA, CCLE, Sanger, GSK, GEO, GTEx, and other public sources. We show how CELLX supports precision oncology through indications discovery, biomarker evaluation, and cell line screening analysis.

Statistical machine learning methods, especially nonparametric Bayesian methods, have become increasingly popular to infer clonal population structure of tumors. Here we describe the treeCRP, an extension of the Chinese restaurant process (CRP), a popular construction used in nonparametric mixture models, to infer the phylogeny and genotype of major subclonal lineages represented in the population of cancer cells. We also propose new split-merge updates tailored to the subclonal reconstruction problem that improve the mixing time of Markov chains. In comparisons with the tree-structured stick breaking prior used in PhyloSub, we demonstrate superior mixing and running time using the treeCRP with our new split-merge procedures. We also show that given the same number of samples, TSSB and treeCRP have similar ability to recover the subclonal structure of a tumor…

Complex mechanisms involving genomic aberrations in numerous proteins and pathways are believed to be a key cause of many diseases such as cancer. With recent advances in genomics, elucidating the molecular basis of cancer at a patient level is now feasible, and has led to personalized treatment strategies whereby a patient is treated according to his or her genomic profile. However, there is growing recognition that existing treatment modalities are overly simplistic, and do not fully account for the deep genomic complexity associated with sensitivity or resistance to cancer therapies. To overcome these limitations, large-scale pharmacogenomic screens of cancer cell lines – in conjunction with modern statistical learning approaches - have been used to explore the genetic underpinnings of drug response. While these analyses have demonstrated the ability to infer genetic predictors of compound sensitivity, to date most modeling approaches have been data-driven, i.e. they do not explicitly incorporate domain-specific knowledge (priors) in the process of learning a model. While a purely data-driven approach offers an unbiased perspective of the data – and may yield unexpected or novel insights - this strategy introduces challenges for both model interpretability and accuracy. In this study, we propose a novel prior-incorporated sparse regression model in which the choice of informative predictor sets is carried out by knowledge-driven priors (gene sets) in a stepwise fashion. Under regularization in a linear regression model, our algorithm is able to incorporate prior biological knowledge across the predictive variables thereby improving the interpretability of the final model with no loss – and often an improvement - in predictive performance. We evaluate the performance of our algorithm compared to well-known regularization methods such as LASSO, Ridge and Elastic net regression in the Cancer Cell Line Encyclopedia (CCLE) and Genomics of Drug Sensitivity in Cancer (Sanger) pharmacogenomics datasets, demonstrating that incorporation of the biological priors selected by our model confers improved predictability and interpretability, despite much fewer predictors, over existing state-of-the-art methods.

We present a genome-wide analysis of splicing patterns of 282 kidney renal clear cell carcinoma patients in which we integrate data from whole-exome sequencing of tumor and normal samples, RNA-seq and copy number variation. We proposed a scoring mechanism to compare splicing patterns in tumor samples to normal samples in order to rank and detect tumor-specific isoforms that have a potential for new biomarkers. We identified a subset of genes that show introns only observable in tumor but not in normal samples, ENCODE and GEUVADIS samples. In order to improve our understanding of the underlying genetic mechanisms of splicing variation we performed a large-scale association analysis to find links between somatic or germline variants with alternative splicing events. We identified 915 cis- and trans-splicing quantitative trait loci (sQTL) associated with changes in splicing patterns. Some of these sQTL have previously been associated with being susceptibility loci for cancer and other diseases. Our analysis also allowed us to identify the function of several COSMIC variants showing significant association with changes in alternative splicing. This demonstrates the potential significance of variants affecting alternative splicing events and yields insights into the mechanisms related to an array of disease phenotypes.

The ability to rapidly sequence the tumor and germline DNA of an individual holds the eventual promise of revolutionizing our ability to match targeted therapies to tumors harboring the associated genetic biomarkers. Analyzing high throughput genomic data consisting of millions of base pairs and discovering alterations in clinically actionable genes in a structured and real time manner is at the crux of personalized testing. This requires a computational architecture that can monitor and track a system within a regulated environment as terabytes of data are reduced to a small number of therapeutically relevant variants, delivered as a diagnostic laboratory developed test. These high complexity assays require data structures that enable real-time and retrospective ad-hoc analysis, with a capability of updating to keep up with the rapidly changing genomic and therapeutic options, all under a regulated environment that is relevant under both CMS and FDA depending on application. We describe a flexible computational framework that uses a paired tumor/normal sample allowing for complete analysis and reporting in approximately 24 hours, providing identification of single nucleotide changes, small insertions and deletions, chromosomal rearrangements, gene fusions and gene expression with positive predictive values over 90%. In this paper we present the challenges in integrating clinical, genomic and annotation databases to provide interpreted draft reports which we utilize within ongoing clinical research protocols. We demonstrate the need to retire from existing performance measurements of accuracy and specificity and measure metrics that are meaningful to a genomic diagnostic environment. This paper presents a three-tier infrastructure that is currently being used to analyze an individual genome and provide available therapeutic options via a clinical report. Our framework utilizes a non-relational variant-centric database that is scaleable to a large amount of data and addresses the challenges and limitations of a relational database system. Our system is continuously monitored via multiple trackers each catering differently to the diversity of users involved in this process. These trackers designed in analytics web-app framework provide status updates for an individual sample accurate to a few minutes. In this paper, we also present our outcome delivery process that is designed and delivered adhering to the standards defined by various regulation agencies involved in clinical genomic testing.

Within the past few decades, drug combination therapy has been intensively studied in oncology and other complex disease areas, especially during the early drug discovery stage, as drug combinations have the potential to improve treatment response, minimize development of resistance or minimize adverse events. In the present, designing combination trials relies mainly on clinical and empirical experience. While empirical experience has indeed crafted efficacious combination therapy clinical trials (combination trials), however, garnering experience with patients can take a lifetime. The preliminary step to eliminating this barrier of time, then, is to understand the current state of combination trials. Thus, we present the first large-scale study of clinical trials (2008-2013) from ClinicalTrials.gov to compare combination trials to non-combination trials, with a focus on oncology. In this work, we developed a classifier to identify combination trials and oncology trials through natural language processing techniques. After clustering trials, we categorized them based on selected characteristics and observed trends present. Among the characteristics studied were primary purpose, funding source, endpoint measurement, allocation, and trial phase. We observe a higher prevalence of combination therapy in oncology (25.6% use combination trials) in comparison to other disease trials (6.9%). However, surprisingly the prevalence of combinations does not increase over the years. In addition, the trials supported by the NIH are significantly more likely to use combinations of drugs than those supported by industry. Our preliminary study of current combination trials may facilitate future trial design and move more preclinical combination studies to the clinical trial stage.

There has been great interest and research initiatives in the biomedical community around harnessing “big data”, including data from the literature, high-throughput gene expression experiments, array CGH and high-throughput siRNA and many other types of data to generate novel hypothesis to address the most crucial biomedical questions and aid in the discovery of more effective and improved therapeutic options for the treatment of complex and pervasive diseases such as cancer. Cancer research has progressed rapidly in the last decade with the implementation of high-dimensional genomic technologies. The large amount of data generated over the years has enabled a systems-based approach to uncovering and elucidating the complex signaling networks associated with cancer. However, even though new technologies have advanced our understanding of cancer biology beyond what could be imagined even a decade ago, there still exist unique challenges associated precisely with the amount of data that is now routinely generated from even a single patient. The data must be stored and processed, with novel analysis strategies called for to uncover new insights into cancer biology that are literally hidden in ‘big data’. Interest in taming ‘big data’ through methods and systems to extract, represent, and transform it into knowledge that can effectively be used for reasoning and question answering will only increase over time, enabling scientists to finally use the data for personalized treatment, discovery and validation…

Here we present a method for extracting candidate cancer pathways from tumor ‘omics data while explicitly accounting for diverse consequences of mutations for protein interactions. Disease-causing mutations are frequently observed at either core or interface residues mediating protein interactions. Mutations at core residues frequently destabilize protein structure while mutations at interface residues can specifically affect the binding energies of protein-protein interactions. As a result, mutations in a protein may result in distinct interaction profiles and thus have different phenotypic consequences. We describe a protein structure-guided pipeline for extracting interacting protein sets specific to a particular mutation. Of 59 cancer genes with 3D co-complexed structures in the Protein Data Bank, 43 showed evidence of mutations with different functional consequences. Literature survey reciprocated functional predictions specific to distinct mutations on APC, ATRX, BRCA1, CBL and HRAS. Our analysis suggests that accounting for mutation-specific perturbations to cancer pathways will be essential for personalized cancer therapy.

Enormous efforts of whole exome and genome sequencing from hundreds to thousands of patients have provided the landscape of somatic genomic alterations in many cancer types to distinguish between driver mutations and passenger mutations. Driver mutations show strong associations with cancer clinical outcomes such as survival. However, due to the heterogeneity of tumors, somatic mutation profiles are exceptionally sparse whereas other types of genomic data such as miRNA or gene expression contain much more complete data for all genomic features with quantitative values measured in each patient. To overcome the extreme sparseness of somatic mutation profiles and allow for the discovery of combinations of somatic mutations that may predict cancer clinical outcomes, here we propose a new approach for binning somatic mutations based on existing biological knowledge. Through the analysis using renal cell carcinoma dataset from The Cancer Genome Atlas (TCGA), we identified combinations of somatic mutation burden based on pathways, protein families, evolutionary conversed regions, and regulatory regions associated with survival. Due to the nature of heterogeneity in cancer, using a binning strategy for somatic mutation profiles based on biological knowledge will be valuable for improved prognostic biomarkers and potentially for tailoring therapeutic strategies by identifying combinations of driver mutations.

We present a new method for exploring cancer gene expression data based on tools from algebraic topology. Our method selects a small relevant subset from tens of thousands of genes while simultaneously identifying nontrivial higher order topological features, i.e., holes, in the data. We first circumvent the problem of high dimensionality by dualizing the data, i.e., by studying genes as points in the sample space. Then we select a small subset of the genes as landmarks to construct topological structures that capture persistent, i.e., topologically significant, features of the data set in its first homology group. Furthermore, we demonstrate that many members of these loops have been implicated for cancer biogenesis in scientific literature. We illustrate our method on five different data sets belonging to brain, breast, leukemia, and ovarian cancers.

Biological pathways are central to understanding complex diseases such as cancer. The majority of this knowledge is scattered in the vast and rapidly growing research literature. To automate knowledge extraction, machine learning approaches typically require annotated examples, which are expensive and time-consuming to acquire. Recently, there has been increasing interest in leveraging databases for distant supervision in knowledge extraction, but existing applications focus almost exclusively on newswire domains. In this paper, we present the first attempt to formulate the distant supervision problem for pathway extraction and apply a state-of-the-art method to extracting pathway interactions from PubMed abstracts. Experiments show that distant supervision can effectively compensate for the lack of annotation, attaining an accuracy approaching supervised results. From 22 million PubMed abstracts, we extracted 1.5 million pathway interactions at a precision of 25%. More than 10% of interactions are mentioned in the context of one or more cancer types, analysis of which yields interesting insights.

Big data bring new opportunities for methods that efficiently summarize and automatically extract knowledge from such compendia. While both supervised learning algorithms and unsupervised clustering algorithms have been successfully applied to biological data, they are either dependent on known biology or limited to discerning the most significant signals in the data. Here we present denoising autoencoders (DAs), which employ a data-defined learning objective independent of known biology, as a method to identify and extract complex patterns from genomic data. We evaluate the performance of DAs by applying them to a large collection of breast cancer gene expression data. Results show that DAs successfully construct features that contain both clinical and molecular information. There are features that represent tumor or normal samples, estrogen receptor (ER) status, and molecular subtypes. Features constructed by the autoencoder generalize to an independent dataset collected using a distinct experimental platform. By integrating data from ENCODE for feature interpretation, we discover a feature representing ER status through association with key transcription factors in breast cancer. We also identify a feature highly predictive of patient survival and it is enriched by FOXM1 signaling pathway. The features constructed by DAs are often bimodally distributed with one peak near zero and another near one, which facilitates discretization. In summary, we demonstrate that DAs effectively extract key biological principles from gene expression data and summarize them into constructed features with convenient properties.

Brain tumor is a fatal central nervous system disease that occurs in around 250,000 people each year globally and it is the second cause of cancer in children. It has been widely acknowledged that genetic factor is one of the significant risk factors for brain cancer. Thus, accurate descriptions of the locations of where the relative genes are active and how these genes express are critical for understanding the pathogenesis of brain tumor and for early detection. The Allen Developing Mouse Brain Atlas is a project on gene expression over the course of mouse brain development stages. Utilizing mouse models allows us to use a relatively homogeneous system to reveal the genetic risk factor of brain cancer. In the Allen atlas, about 435,000 high-resolution spatiotemporal in situ hybridization images have been generated for approximately 2,100 genes and currently the expression patterns over specific brain regions are manually annotated by experts, which does not scale with the continuously expanding collection of images. In this paper, we present an efficient computational approach to perform automated gene expression pattern annotation on brain images. First, the gene expression information in the brain images is captured by invariant features extracted from local image patches. Next, we adopt an augmented sparse coding method, called Stochastic Coordinate Coding, to construct high-level representations. Different pooling methods are then applied to generate gene-level features. To discriminate gene expression patterns at specific brain regions, we employ supervised learning methods to build accurate models for both binary-class and multi-class cases. Random undersampling and majority voting strategies are utilized to deal with the inherently imbalanced class distribution within each annotation task in order to further improve predictive performance. In addition, we propose a novel structure-based multi-label classification approach, which makes use of label hierarchy based on brain ontology during model learning. Extensive experiments have been conducted on the atlas and results show that the proposed approach produces higher annotation accuracy than several baseline methods. Our approach is shown to be robust on both binary-class and multi-class tasks and even with a relatively low training ratio. Our results also show that the use of label hierarchy can significantly improve the annotation accuracy at all brain ontology levels.

While genome-wide association studies (GWAS) [1] have identified the genetic underpinning of a number of complex traits, large portions of the heritability of common, complex diseases are still unknown [2-6]. Beyond the association between genetic variation and outcomes, the impact of environment exposure, as well as gene-gene (G×G) and gene-environment (G×E) interactions, are undoubtedly fundamental mechanisms involved in the development of complex traits. Novel methods tailored to detect these predictors have the potential to (1) reveal the impact of multiple variations in biological pathways and (2) identify genes that are only associated with a particular disease in the presence of a given environmental exposure (e.g. smoking). Such knowledge could be used to assess personal risk and to choose suitable medical interventions, based on an individual’s genotype and environmental exposures. Further, a more complete picture of the genetic and environmental aspects that impact complex disease can be used to inform environmental regulations to protect vulnerable populations…

Studies assessing the impact of gene-environment interactions on common human diseases and traits have been relatively few for many reasons. One often acknowledged reason is that it is difficult to accurately measure the environment or exposure. Indeed, most large-scale epidemiologic studies use questionnaires to assess and measure past and current exposure levels. While questionnaires may be cost-effective, the data may or may not accurately represent the exposure compared with more direct measurements (e.g., self-reported current smoking status versus direct measurement for cotinine levels). Much like phenotyping, the choice in how an exposure is measured may impact downstream tests of genetic association and gene-environment interaction studies. As a case study, we performed tests of association between five common VKORC1 SNPs and two different measurements of vitamin K levels, dietary (n=5,725) and serum (n=348), in the Third National Health and Nutrition Examination Studies (NHANES III). We did not replicate previously reported associations between VKORC1 and vitamin K levels using either measure. Furthermore, the suggestive associations and estimated genetic effect sizes identified in this study differed depending on the vitamin K measurement. This case study of VKORC1 and vitamin K levels serves as a cautionary example of the downstream consequences that the type of exposure measurement choices will have on genetic association and possibly gene-environment studies.

Environmental exposure is a key factor of understanding health and diseases. Beyond genetic propensities, many disorders are, in part, caused by human interaction with harmful substances in the water, the soil, or the air. Limited data is available on a disease or substance basis. However, we compile a global repository from literature surveys matching environmental chemical substances exposure with human disorders. We build a bipartite network linking 60 substances to over 150 disease phenotypes. We quantitatively and qualitatively analyze the network and its projections as simple networks. We identify mercury, lead and cadmium as associated with the largest number of disorders. Symmetrically, we show that breast cancer, harm to the fetus and non-Hodgkin’s lymphoma are associated with the most environmental chemicals. We conduct statistical analysis of how vertices with similar characteristics form the network interactions. This dyadicity and heterophilicity measures the tendencies of vertices with similar properties to either connect to one-another. We study the dyadic distribution of the substance classes in the networks show that, for instance, tobacco smoke compounds, parabens and heavy metals tend to be connected, which hint at common disease causing factors, whereas fungicides and phytoestrogens do not. We build an exposure network at the systems level. The information gathered in this study is meant to be complementary to the genome and help us understand complex diseases, their commonalities, their causes, and how to prevent and treat them.

Gene-environment (G×E) interactions are biologically important for a wide range of environmental exposures and clinical outcomes. Because of the large number of potential interactions in genomewide association data, the standard approach fits one model per G×E interaction with multiple hypothesis correction (MHC) used to control the type I error rate. Although sometimes effective, using one model per candidate G×E interaction test has two important limitations: low power due to MHC and omitted variable bias. To avoid the coefficient estimation bias associated with independent models, researchers have used penalized regression methods to jointly test all main effects and interactions in a single regression model. Although penalized regression supports joint analysis of all interactions, can be used with hierarchical constraints, and offers excellent predictive performance, it cannot assess the statistical significance of G×E interactions or compute meaningful estimates of effect size. To address the challenge of low power, researchers have separately explored screening-testing, or two-stage, methods in which the set of potential G×E interactions is first filtered and then tested for interactions with MHC only applied to the tests actually performed in the second stage. Although two-stage methods are statistically valid and effective at improving power, they still test multiple separate models and so are impacted by MHC and biased coefficient estimation. To remedy the challenges of both poor power and omitted variable bias encountered with traditional G×E interaction detection methods, we propose a novel approach that combines elements of screening-testing and hierarchical penalized regression. Specifically, our proposed method uses, in the first stage, an elastic net-penalized multiple logistic regression model to jointly estimate either the marginal association filter statistic or the gene-environment correlation filter statistic for all candidate genetic markers. In the second stage, a single multiple logistic regression model is used to jointly assess marginal terms and G×E interactions for all genetic markers that pass the first stage filter. A single likelihood-ratio test is used to determine whether any of the interactions are statistically significant. We demonstrate the efficacy of our method relative to alternative G×E detection methods on a bladder cancer data set.

Standard analysis methods for genome wide association studies (GWAS) are not robust to complex disease models, such as interactions between variables with small main effects. These types of effects likely contribute to the heritability of complex human traits. Machine learning methods that are capable of identifying interactions, such as Random Forests (RF), are an alternative analysis approach. One caveat to RF is that there is no standardized method of selecting variables so that false positives are reduced while retaining adequate power. To this end, we have developed a novel variable selection method called relative recurrency variable importance metric (r2VIM). This method incorporates recurrency and variance estimation to assist in optimal threshold selection. For this study, we specifically address how this method performs in data with almost completely epistatic effects (i.e. no marginal effects). Our results show that with appropriate parameter settings, r2VIM can identify interaction effects when the marginal effects are virtually nonexistent. It also outperforms logistic regression, which has essentially no power under this type of model when the number of potential features (genetic variants) is large. (All Supplementary Data can be found here: http://research.nhgri.nih.gov/manuscripts/Bailey-Wilson/r2VIM_epi/).

The large volume of GWAS data poses great computational challenges for analyzing genetic interactions associated with common human diseases. We propose a computational framework for characterizing epistatic interactions among large sets of genetic attributes in GWAS data. We build the human phenotype network (HPN) and focus around a disease of interest. In this study, we use the GLAUGEN glaucoma GWAS dataset and apply the HPN as a biological knowledge-based filter to prioritize genetic variants. Then, we use the statistical epistasis network (SEN) to identify a significant connected network of pairwise epistatic interactions among the prioritized SNPs. These clearly highlight the complex genetic basis of glaucoma. Furthermore, we identify key SNPs by quantifying structural network characteristics. Through functional annotation of these key SNPs using Biofilter, a software accessing multiple publicly available human genetic data sources, we find supporting biomedical evidences linking glaucoma to an array of genetic diseases, proving our concept. We conclude by suggesting hypotheses for a better understanding of the disease.

Elevated levels of plasma fibrinogen are associated with clot formation in the absence of inflammation or injury and is a biomarker for arterial clotting, the leading cause of cardiovascular disease. Fibrinogen levels are heritable with >50% attributed to genetic factors, however little is known about possible genetic modifiers that might explain the missing heritability. The fibrinogen gene cluster is comprised of three genes (FGA, FGB, and FGG) that make up the fibrinogen polypeptide essential for fibrinogen production in the blood. Given the known interaction with these genes, we tested 25 variants in the fibrinogen gene cluster for gene x gene and gene x environment interactions in 620 non-Hispanic blacks, 1,385 non-Hispanic whites, and 664 Mexican Americans from a cross-sectional dataset enriched with environmental data, the Third National Health and Nutrition Examination Survey (NHANES III). Using a multiplicative approach, we added cross product terms (gene x gene or gene x environment) to a linear regression model and declared significance at p < 0.05. We identified 19 unique gene x gene and 13 unique gene x environment interactions that impact fibrinogen levels in at least one population at p <0.05. Over 90% of the gene x gene interactions identified include a variant in the rate-limiting gene, FGB that is essential for the formation of the fibrinogen polypeptide. We also detected gene x environment interactions with fibrinogen variants and sex, smoking, and body mass index. These findings highlight the potential for the discovery of genetic modifiers for complex phenotypes in multiple populations and give a better understanding of the interaction between genes and/or the environment for fibrinogen levels. The need for more powerful and robust methods to identify genetic modifiers is still warranted.

The environment plays a major role in influencing diseases and health. The phenomenon of environmental exposure is complex and humans are not exposed to one or a handful factors but potentially hundreds factors throughout their lives. The exposome, the totality of exposures encountered from birth, is hypothesized to consist of multiple inter-dependencies, or correlations, between individual exposures. These correlations may reflect how individuals are exposed. Currently, we lack methods to comprehensively identify robust and replicated correlations between environmental exposures of the exposome. Further, we have not mapped how exposures associated with disease identified by environment-wide association studies (EWAS) are correlated with other exposures. To this end, we implement methods to describe a first “exposome globe”, a comprehensive display of replicated correlations between individual exposures of the exposome. First, we describe overall characteristics of the dense correlations between exposures, showing that we are able to replicate 2,656 correlations between individual exposures of 81,937 total considered (3%). We document the correlation within and between broad a priori defined categories of exposures (e.g., pollutants and nutrient exposures). We also demonstrate utility of the exposome globe to contextualize exposures found through two EWASs in type 2 diabetes and all-cause mortality, such as exposure clusters putatively related to smoking behaviors and persistent pollutant exposure. The exposome globe construct is a useful tool for the display and communication of the complex relationships between exposure factors and between exposure factors related to disease status.

Substantial progress has been made in identifying susceptibility variants for age-related macular degeneration (AMD). The majority of research to identify genetic variants associated with AMD has focused on nuclear genetic variation. While there is some evidence that mitochondrial genetic variation contributes to AMD susceptibility, to date, these studies have been limited to populations of European descent resulting in a lack of data in diverse populations. A major goal of the Epidemiologic Architecture for Genes Linked to Environment (EAGLE) study is to describe the underlying genetic architecture of common, complex diseases across diverse populations. This present study sought to determine if mitochondrial genetic variation influences risk of AMD across diverse populations. We performed a genetic association study to investigate the contribution of mitochondrial DNA variation to AMD risk. We accessed samples from the National Health and Nutrition Examination Surveys, a U.S population-based, cross-sectional survey collected without regard to health status. AMD cases and controls were selected from the Third NHANES and NHANES 2007-2008 datasets which include non-Hispanic whites, non-Hispanic blacks, and Mexican Americans. AMD cases were defined as those > 60 years of age with early/late AMD, as determined by fundus photography. Targeted genotyping was performed for 63 mitochondrial SNPs and participants were then classified into mitochondrial haplogroups. We used logistic regression assuming a dominant genetic model adjusting for age, sex, body mass index, and smoking status (ever vs. never). Regressions and meta-analyses were performed for individual SNPs and mitochondrial haplogroups J, T, and U. We identified five SNPs associated with AMD in Mexican Americans at p < 0.05, including three located in the control region (mt16111, mt16362, and mt16319), one in MT-RNR2 (mt1736), and one in MT-ND4 (mt12007). No mitochondrial variant or haplogroup was significantly associated in non-Hispanic blacks or non- Hispanic whites in the final meta-analysis. This study provides further evidence that mitochondrial variation plays a role in susceptibility to AMD and contributes to the knowledge of the genetic architecture of AMD in Mexican Americans.

We introduce the integrative protein-interaction-network-based pathway analysis (iPINBPA) for genome-wide association studies (GWAS), a method to identify and prioritize genetic associations by merging statistical evidence of association with physical evidence of interaction at the protein level. First, the strongest associations are used to weight all nodes in the PPI network using a guilt- by-association approach. Second, the gene-wise converted p-values from a GWAS are integrated with node weights using the Liptak-Stouffer method. Finally, a greedy search is performed to find enriched modules, i.e., sub-networks with nodes that have low p-values and high weights. The performance of iPINBPA and other state-of-the-art methods is assessed by computing the concentrated receiver operating characteristic (CROC) curves using two independent multiple sclerosis (MS) GWAS studies and one recent ImmunoChip study. Our results showed that iPINBPA identified sub-networks with smaller sizes and higher enrichments than other methods. iPINBPA offers a novel strategy to integrate topological connectivity and association signals from GWAS, making this an attractive tool to use in other large GWAS datasets.

The following sections are included:

• Introduction
• Session articles
• Acknowledgments
• References

A pilot reputation-based collaborative network biology platform, Bionet, was developed for use in the sbv IMPROVER Network Verification Challenge to verify and enhance previously developed networks describing key aspects of lung biology. Bionet was successful in capturing a more comprehensive view of the biology associated with each network using the collective intelligence and knowledge of the crowd. One key learning point from the pilot was that using a standardized biological knowledge representation language such as BEL is critical to the success of a collaborative network biology platform. Overall, Bionet demonstrated that this approach to collaborative network biology is highly viable. Improving this platform for de novo creation of biological networks and network curation with the suggested enhancements for scalability will serve both academic and industry systems biology communities.

Identifying concepts and relationships in biomedical text enables knowledge to be applied in computational analyses. Many biological natural language processing (BioNLP) projects attempt to address this challenge, but the state of the art still leaves much room for improvement. Progress in BioNLP research depends on large, annotated corpora for evaluating information extraction systems and training machine learning models. Traditionally, such corpora are created by small numbers of expert annotators often working over extended periods of time. Recent studies have shown that workers on microtask crowdsourcing platforms such as Amazon’s Mechanical Turk (AMT) can, in aggregate, generate high-quality annotations of biomedical text. Here, we investigated the use of the AMT in capturing disease mentions in PubMed abstracts. We used the NCBI Disease corpus as a gold standard for refining and benchmarking our crowdsourcing protocol. After several iterations, we arrived at a protocol that reproduced the annotations of the 593 documents in the ‘training set’ of this gold standard with an overall F measure of 0.872 (precision 0.862, recall 0.883). The output can also be tuned to optimize for precision (max = 0.984 when recall = 0.269) or recall (max = 0.980 when precision = 0.436). Each document was completed by 15 workers, and their annotations were merged based on a simple voting method. In total 145 workers combined to complete all 593 documents in the span of 9 days at a cost of $.066 per abstract per worker. The quality of the annotations, as judged with the F measure, increases with the number of workers assigned to each task; however minimal performance gains were observed beyond 8 workers per task. These results add further evidence that microtask crowdsourcing can be a valuable tool for generating well-annotated corpora in BioNLP. Data produced for this analysis are available at http://figshare.com/articles/Disease_Mention_Annotation_with_Mechanical_Turk/1126402.

The development of tools in computational pathology to assist physicians and biomedical scientists in the diagnosis of disease requires access to high-quality annotated images for algorithm learning and evaluation. Generating high-quality expert-derived annotations is time-consuming and expensive. We explore the use of crowdsourcing for rapidly obtaining annotations for two core tasks in com- putational pathology: nucleus detection and nucleus segmentation. We designed and implemented crowdsourcing experiments using the CrowdFlower platform, which provides access to a large set of labor channel partners that accesses and manages millions of contributors worldwide. We obtained annotations from four types of annotators and compared concordance across these groups. We obtained: crowdsourced annotations for nucleus detection and segmentation on a total of 810 images; annotations using automated methods on 810 images; annotations from research fellows for detection and segmentation on 477 and 455 images, respectively; and expert pathologist-derived annotations for detection and segmentation on 80 and 63 images, respectively. For the crowdsourced annotations, we evaluated performance across a range of contributor skill levels (1, 2, or 3). The crowdsourced annotations (4,860 images in total) were completed in only a fraction of the time and cost required for obtaining annotations using traditional methods. For the nucleus detection task, the research fellow-derived annotations showed the strongest concordance with the expert pathologist- derived annotations (F-M =93.68%), followed by the crowd-sourced contributor levels 1,2, and 3 and the automated method, which showed relatively similar performance (F-M = 87.84%, 88.49%, 87.26%, and 86.99%, respectively). For the nucleus segmentation task, the crowdsourced contributor level 3-derived annotations, research fellow-derived annotations, and automated method showed the strongest concordance with the expert pathologist-derived annotations (F-M = 66.41%, 65.93%, and 65.36%, respectively), followed by the contributor levels 2 and 1 (60.89% and 60.87%, respectively). When the research fellows were used as a gold-standard for the segmentation task, all three con- tributor levels of the crowdsourced annotations significantly outperformed the automated method (F-M = 62.21%, 62.47%, and 65.15% vs. 51.92%). Aggregating multiple annotations from the crowd to obtain a consensus annotation resulted in the strongest performance for the crowd-sourced segmentation. For both detection and segmentation, crowd-sourced performance is strongest with small images (400 × 400 pixels) and degrades significantly with the use of larger images (600 × 600 and 800 × 800 pixels). We conclude that crowdsourcing to non-experts can be used for large-scale labeling microtasks in computational pathology and offers a new approach for the rapid generation of labeled images for algorithm development and evaluation.

Post-market drug safety surveillance is hugely important and is a significant challenge despite the existence of adverse event (AE) reporting systems. Here we describe a preliminary analysis of search logs from healthcare professionals as a source for detecting adverse drug events. We annotate search log query terms with biomedical terminologies for drugs and events, and then perform a statistical analysis to identify associations among drugs and events within search sessions. We evaluate our approach using two different types of reference standards consisting of known adverse drug events (ADEs) and negative controls. Our approach achieves a discrimination accuracy of 0.85 in terms of the area under the receiver operator curve (AUC) for the reference set of well-established ADEs and an AUC of 0.68 for the reference set of recently labeled ADEs. We also find that the majority of associations in the reference sets have support in the search log data. Despite these promising results additional research is required to better understand users’ search behavior, biasing factors, and the overall utility of analyzing healthcare professional search logs for drug safety surveillance.

The availability of high-quality physical interaction datasets is a prerequisite for system-level analysis of interactomes and supervised models to predict protein-protein interactions (PPIs). One source is literature-curated PPI databases in which pairwise associations of proteins published in the scientific literature are deposited. However, PPIs may not be clearly labelled as physical interactions affecting the quality of the entire dataset. In order to obtain a high-quality gold standard dataset for PPIs between human immunodeficiency virus (HIV-1) and its human host, we adopted a crowd-sourcing approach. We collected expert opinions and utilized an expectation-maximization based approach to estimate expert labeling quality. These estimates are used to infer the probability of a reported PPI actually being a direct physical interaction given the set of expert opinions. The effectiveness of our approach is demonstrated through synthetic data experiments and a high quality physical interaction network between HIV and human proteins is obtained. Since many literature-curated databases suffer from similar challenges, the framework described herein could be utilized in refining other databases. The curated data is available at http://www.cs.bilkent.edu.tr/~oznur.tastan/supp/psb2015/…

The annotation and classification of ncRNAs is essential to decipher molecular mechanisms of gene regulation in normal and disease states. A database such as Rfam maintains alignments, consensus secondary structures, and corresponding annotations for RNA families. Its primary purpose is the automated, accurate annotation of non-coding RNAs in genomic sequences. However, the alignment of RNAs is computationally challenging, and the data stored in this database are often subject to improvements. Here, we design and evaluate Ribo, a human-computing game that aims to improve the accuracy of RNA alignments already stored in Rfam. We demonstrate the potential of our techniques and discuss the feasibility of large scale collaborative annotation and classification of RNA families.

Advances in molecular profiling and sensor technologies are expanding the scope of personalized medicine beyond genotypes, providing new opportunities for developing richer and more dynamic multi-scale models of individual health. Recent studies demonstrate the value of scoring high-dimensional microbiome, immune, and metabolic traits from individuals to inform personalized medicine. Efforts to integrate multiple dimensions of clinical and molecular data towards predictive multi-scale models of individual health and wellness are already underway. Improved methods for mining and discovery of clinical phenotypes from electronic medical records and technological developments in wearable sensor technologies present new opportunities for mapping and exploring the critical yet poorly characterized “phenome” and “envirome” dimensions of personalized medicine. There are ambitious new projects underway to collect multi-scale molecular, sensor, clinical, behavioral, and environmental data streams from large population cohorts longitudinally to enable more comprehensive and dynamic models of individual biology and personalized health. Personalized medicine stands to benefit from inclusion of rich new sources and dimensions of data. However, realizing these improvements in care relies upon novel informatics methodologies, tools, and systems to make full use of these data to advance both the science and translational applications of personalized medicine…

In eukaryotic cells, alternative cleavage of 3’ untranslated regions (UTRs) can affect transcript stability, transport and translation. For polyadenylated (poly(A)) transcripts, cleavage sites can be characterized with short-read sequencing using specialized library construction methods. However, for large-scale cohort studies as well as for clinical sequencing applications, it is desirable to characterize such events using RNA-seq data, as the latter are already widely applied to identify other relevant information, such as mutations, alternative splicing and chimeric transcripts. Here we describe KLEAT, an analysis tool that uses de novo assembly of RNA-seq data to characterize cleavage sites on 3’ UTRs. We demonstrate the performance of KLEAT on three cell line RNA-seq libraries constructed and sequenced by the ENCODE project, and assembled using Trans-ABySS. Validating the KLEAT predictions with matched ENCODE RNA-seq and RNA-PET libraries, we show that the tool has over 90% positive predictive value when there are at least three RNA-seq reads supporting a poly(A) tail and requiring at least three RNA-PET reads mapping within 100 nucleotides as validation. We also compare the performance of KLEAT with other popular RNA-seq analysis pipelines that reconstruct 3’ UTR ends, and show that it performs favourably, based on an ROC-like curve.

Inferring causal relationships among molecular and higher order phenotypes is a critical step in elucidating the complexity of living systems. Here we propose a novel method for inferring causality that is no longer constrained by the conditional dependency arguments that limit the ability of statistical causal inference methods to resolve causal relationships within sets of graphical models that are Markov equivalent. Our method utilizes Bayesian belief propagation to infer the responses of perturbation events on molecular traits given a hypothesized graph structure. A distance measure between the inferred response distribution and the observed data is defined to assess the ’fitness’ of the hypothesized causal relationships. To test our algorithm, we infer causal relationships within equivalence classes of gene networks in which the form of the functional interactions that are possible are assumed to be nonlinear, given synthetic microarray and RNA sequencing data. We also apply our method to infer causality in real metabolic network with v-structure and feedback loop. We show that our method can recapitulate the causal structure and recover the feedback loop only from steady-state data which conventional method cannot.

The promise of personalized medicine will require rigorously validated molecular diagnostics developed on minimally invasive, clinically relevant samples. Measurement of DNA mutations is increasingly common in clinical settings but only higher-prevalence mutations are cost-effective. Patients with rare variants are at best ignored or, at worst, misdiagnosed. Mutations result in downstream impacts on transcription, offering the possibility of broader diagnosis for patients with rare variants causing similar downstream changes. Use of such signatures in clinical settings is rare as these algorithms are difficult to validate for commercial use. Validation on a test set (against a clinical gold standard) is necessary but not sufficient: accuracy must be maintained amidst interfering substances, across reagent lots and across operators. Here we report the development, clinical validation, and diagnostic accuracy of a pre-operative molecular test (Afirma BRAF) to identify BRAF V600E mutations using mRNA expression in thyroid fine needle aspirate biopsies (FNABs). FNABs were obtained prospectively from 716 nodules and more than 3,000 features measured using microarrays. BRAF V600E labels for training (n=181) and independent test (n=535) sets were established using a sensitive quantitative PCR (qPCR) assay. The resulting 128-gene linear support vector machine was compared to qPCR in the independent test set. Clinical sensitivity and specificity for malignancy were evaluated in a subset of test set samples (n=213) with expert-derived histopathology. We observed high positive- (PPA, 90.4%) and negative (NPA, 99.0%) percent agreement with qPCR on the test set. Clinical sensitivity for malignancy was 43.8% (consistent with published prevalence of BRAF V600E in this neoplasm) and specificity was 100%, identical to qPCR on the same samples. Classification was accurate in up to 60% blood. A double-mutant still resulting in the V600E amino acid change was negative by qPCR but correctly positive by Afirma BRAF. Non-diagnostic rates were lower (7.6%) for Afirma BRAF than for qPCR (24.5%), a further advantage of using RNA in small sample biopsies. Afirma BRAF accurately determined the presence or absence of the BRAF V600E DNA mutation in FNABs, a collection method directly relevant to solid tumor assessment, with performance equal to that of an established, highly sensitive DNA-based assay and with a lower non-diagnostic rate. This is the first such test in thyroid cancer to undergo sufficient analytical and clinical validation for real-world use in a personalized medicine context to frame individual patient risk and inform surgical choice.

Gene expression and disease-associated variants are often used to prioritize candidate genes for target validation. However, the success of these gene features alone or in combination in the discovery of therapeutic targets is uncertain. Here we evaluated the effectiveness of the differential expression (DE), the disease-associated single nucleotide polymorphisms (SNPs) and the combination of the two in recovering and predicting known therapeutic targets across 56 human diseases. We demonstrate that the performance of each feature varies across diseases and generally the features have more recovery power than predictive power. The combination of the two features, however, has significantly higher predictive power than each feature alone. Our study provides a systematic evaluation of two common gene features, DE and SNPs, for prioritization of candidate targets and identified an improved predictive power of coupling these two features.

The blood gene expression signatures are used as biomarkers for immunological and non- immunological diseases. Therefore, it is important to understand the variation in blood gene expression patterns and the factors (heritable/non-heritable) that underlie this variation. In this paper, we study the relationship between drug effects on the one hand, and heritable and non-heritable factors influencing gene expression on the other. Understanding of this relationship can help select appropriate targets for drugs aimed at reverting disease phenotypes to healthy states. In order to estimate heritable and non-heritable effects on gene expression, we use Twin-ACE model on a gene expression dataset MuTHER, measured in blood samples from monozygotic and dizygotic twins. In order to associate gene expression with drug effects, we use CMap database. We show that, even though the expressions of most genes are driven by non-heritable factors, drugs are more likely to influence expression of genes, driven by heritable rather than non-heritable factors. We further study this finding in the context of a gene regulatory network. We investigate the relationship between the drug effects on gene expression and propagation of heritable and non-heritable factors through regulatory networks. We find that the decisive factor in determining whether a gene will be influenced by a drug is the flow of heritable effects supplied to the gene through regulatory network.

In the past decade there has been an explosion in genetic research that has resulted in the generation of enormous quantities of disease-related data. In the current study, we have compiled disease risk gene variant information and Electronic Medical Record (EMR) classification codes from various repositories for 305 diseases. Using such data, we developed a pipeline to test for clinical prevalence, gene-variant overlap, and literature presence for all 46,360 unique diseases pairs. To determine whether disease pairs were enriched we systematically employed both Fishers’ Exact (medical and literature) and Term Frequency-Inverse Document Frequency (genetics) methodologies to test for enrichment, defining statistical significance at a Bonferonni adjusted threshold of (p < 1×10⁻⁶) and weighted q<0.05 accordingly. We hypothesize that disease pairs that are statistically enriched in medical and genetic spheres, but not so in the literature have the potential to reveal non-obvious connections between clinically disparate phenotypes. Using this pipeline, we identified 2,316 disease pairs that were significantly enriched within an EMR and 213 enriched genetically. Of these, 65 disease pairs were statistically enriched in both, 19 of which are believed to be novel. These identified non-obvious relationships between disease pairs are suggestive of a shared underlying etiology with clinical presentation. Further investigation of uncovered disease-pair relationships has the potential to provide insights into the architecture of complex diseases, and update existing knowledge of risk factors.

Increasing availability of high-dimensional clinical data, which improves the ability to define more specific phenotypes, as well as molecular data, which can elucidate disease mechanisms, is a driving force and at the same time a major challenge for translational and personalized medicine. Successful research in this field requires an approach that ties together specific disease and health expertise with understanding of molecular data through statistical methods. We present PEAX (Phenotype-Expression Association eXplorer), built upon open-source software, which integrates visual phenotype model definition with statistical testing of expression data presented concurrently in a web-browser. The integration of data and analysis tasks in a single tool allows clinical domain experts to obtain new insights directly through exploration of relationships between multivariate phenotype models and gene expression data, showing the effects of model definition and modification while also exploiting potential meaningful associations between phenotype and miRNA-mRNA regulatory relationships. We combine the web visualization capabilities of Shiny and D3 with the power and speed of R for backend statistical analysis, in order to abstract the scripting required for repetitive analysis of sub-phenotype association. We describe the motivation for PEAX, demonstrate its utility through a use case involving heart failure research, and discuss computational challenges and observations. We show that our visual web-based representations are well-suited for rapid exploration of phenotype and gene expression association, facilitating insight and discovery by domain experts.

Feature selection is used extensively in biomedical research for biomarker identification and patient classification, both of which are essential steps in developing personalized medicine strategies. However, the structured nature of the biological datasets and high correlation of variables frequently yield multiple equally optimal signatures, thus making traditional feature selection methods unstable. Features selected based on one cohort of patients, may not work as well in another cohort. In addition, biologically important features may be missed due to selection of other co-clustered features We propose a new method, Tree-guided Recursive Cluster Selection (T-ReCS), for efficient selection of grouped features. T-ReCS significantly improves predictive stability while maintains the same level of accuracy. T-ReCS does not require an a priori knowledge of the clusters like group-lasso and also can handle “orphan” features (not belonging to a cluster). T-ReCS can be used with categorical or survival target variables. Tested on simulated and real expression data from breast cancer and lung diseases and survival data, T-ReCS selected stable cluster features without significant loss in classification accuracy.

We present a novel statistical framework for meta-analysis of differential gene co-expression. In contrast to standard methods, which identify genes that are over or under expressed in disease vs controls, differential co-expression identifies gene pairs with correlated expression profiles specific to one state. We apply our differential co-expression meta-analysis method to identify genes specifically mis-expressed in blood-derived cells of systemic lupus erythematosus (SLE) patients. The resulting network is strongly enriched for genes genetically associated with SLE, and effectively identifies gene modules known to play important roles in SLE etiology, such as increased type 1 interferon response and response to wounding. Our results also strongly support previous preliminary studies suggesting a role for dysregulation of neutrophil extracellular trap formation in SLE. Strikingly, two of the gene modules we identify contain SLE-associated transcription factors that have binding sites significantly enriched in the promoter regions of their respective gene modules, suggesting a possible mechanism underlying the mis-expression of the modules. Thus, our general method is capable of identifying specific dysregulated gene expression programs, as opposed to large global responses. We anticipate that methods such as ours will be more and more useful as gene expression monitoring becomes increasingly common in clinical settings.

Recent studies have revealed that melancholic depression, one major subtype of depression, is closely associated with the concentration of some metabolites and biological functions of certain genes and pathways. Meanwhile, recent advances in biotechnologies have allowed us to collect a large amount of genomic data, e.g., metabolites and microarray gene expression. With such a huge amount of information available, one approach that can give us new insights into the understanding of the fundamental biology underlying melancholic depression is to build disease status prediction models using classification or regression methods. However, the existence of strong empirical correlations, e.g., those exhibited by genes sharing the same biological pathway in microarray profiles, tremendously limits the performance of these methods. Furthermore, the occurrence of missing values which are ubiquitous in biomedical applications further complicates the problem. In this paper, we hypothesize that the problem of missing values might in some way benefit from the correlation between the variables and propose a method to learn a compressed set of representative features through an adapted version of sparse coding which is capable of identifying correlated variables and addressing the issue of missing values simultaneously. An efficient algorithm is also developed to solve the proposed formulation. We apply the proposed method on metabolic and microarray profiles collected from a group of subjects consisting of both patients with melancholic depression and healthy controls. Results show that the proposed method can not only produce meaningful clusters of variables but also generate a set of representative features that achieve superior classification performance over those generated by traditional clustering and data imputation techniques. In particular, on both datasets, we found that in comparison with the competing algorithms, the representative features learned by the proposed method give rise to significantly improved sensitivity scores, suggesting that the learned features allow prediction with high accuracy of disease status in those who are diagnosed with melancholic depression. To our best knowledge, this is the first work that applies sparse coding to deal with high feature correlations and missing values, which are common challenges in many biomedical applications. The proposed method can be readily adapted to other biomedical applications involving incomplete and high-dimensional data.

In this paper, we present a novel feature allocation model to describe tumor heterogeneity (TH) using next-generation sequencing (NGS) data. Taking a Bayesian approach, we extend the Indian buffet process (IBP) to define a class of nonparametric models, the categorical IBP (cIBP). A cIBP takes categorical values to denote homozygous or heterozygous genotypes at each SNV. We define a subclone as a vector of these categorical values, each corresponding to an SNV. Instead of partitioning somatic mutations into non-overlapping clusters with similar cellular prevalences, we took a different approach using feature allocation. Importantly, we do not assume somatic mutations with similar cellular prevalence must be from the same subclone and allow overlapping mutations shared across subclones. We argue that this is closer to the underlying theory of phylogenetic clonal expansion, as somatic mutations occurred in parent subclones should be shared across the parent and child subclones. Bayesian inference yields posterior probabilities of the number, genotypes, and proportions of subclones in a tumor sample, thereby providing point estimates as well as variabilities of the estimates for each subclone. We report results on both simulated and real data. BayClone is available at http://health.bsd.uchicago.edu/yji/soft.html.

The following sections are included:

• Introduction
• Speakers and Abstracts

New discoveries in biological, biomedical and health sciences are increasingly being driven by our ability to acquire, share, integrate and analyze, and construct and simulate predictive models of biological systems. While much attention has focused on automating routine aspects of management and analysis of “big data”, realizing the full potential of “big data” to accelerate discovery calls for automating many other aspects of the scientific process that have so far largely resisted automation: identifying gaps in the current state of knowledge; generating and prioritizing questions; designing studies; designing, prioritizing, planning, and executing experiments; interpreting results; forming hypotheses; drawing conclusions; replicating studies; validating claims; documenting studies; communicating results; reviewing results; and integrating results into the larger body of knowledge in a discipline. Against this background, the PSB workshop on Discovery Informatics in Biological and Biomedical Sciences explores the opportunities and challenges of automating discovery or assisting humans in discovery through advances (i) Understanding, formalization, and information processing accounts of, the entire scientific process; (ii) Design, development, and evaluation of the computational artifacts (representations, processes) that embody such understanding; and (iii) Application of the resulting artifacts and systems to advance science (by augmenting individual or collective human efforts, or by fully automating science)…

This workshop will focus on disruptive processes impacting research arising from the increasing ability of individuals to create, curate and share data with scientists. Encompassing processes from funding research to providing samples to creating algorithms, including the public will require new approaches even as it opens up new possibilities. We will hear from a few researchers at the forefront of these disruptive processes, followed by a moderated discussion with the audience about these topics.

The following sections are included:

• Workshop Focus
• Workshop Contributions
• References

Investigating the association between biobank derived genomic data and the information of linked electronic health records (EHRs) is an emerging area of research for dissecting the architecture of complex human traits, where cases and controls for study are defined through the use of electronic phenotyping algorithms deployed in large EHR systems. For our study, cataract cases and controls were identified within the Marshfield Personalized Medicine Research Project (PMRP) biobank and linked EHR, which is a member of the NHGRI-funded electronic Medical Records and Genomics (eMERGE) Network. Our goal was to explore potential gene-gene and gene-environment interactions within these data for 527,953 and 527,936 single nucleotide polymorphisms (SNPs) for gene-gene and gene-environment analyses, respectively, with minor allele frequency > 1%, in order to explore higher level associations with cataract risk beyond investigations of single SNP-phenotype associations. To build our SNP-SNP interaction models we utilized a prior-knowledge driven filtering method called Biofilter to minimize the multiple testing burden of exploring the vast array of interaction models possible from our extensive number of SNPs. Using Biofilter, we developed 57,376 prior-knowledge directed SNP-SNP models to test for association with cataract status. We selected models that required 6 sources of external domain knowledge. We identified 13 statistically significant SNP-SNP models with an interaction with p-value < 1×10⁻⁴, as well as an overall model with p-value < 0.01 associated with cataract status. We also conducted gene-environment interaction analyses for all GWAS SNPs and a set of environmental factors from the PhenX Toolkit: smoking, UV exposure, and alcohol use; these environmental factors have been previously associated with the formation of cataracts. We found a total of 782 gene-environment models that exhibit an interaction with a p-value < 1×10⁻⁴ associated with cataract status. Our results show these approaches enable advanced searches for epistasis and gene-environment interactions beyond GWAS, and that the EHR based approach provides an additional source of data for seeking these advanced explanatory models of the etiology of complex disease/outcome such as cataracts.