Biocomputing 2016

The following sections are included:

• Contents
• Preface
• Thanks to the reviewers…

The following sections are included:

• Introduction
• Overview of Session Contributions
• Discussion and Conclusions
• References

Complex diseases are the result of intricate interactions between genetic, epigenetic and environmental factors. In previous studies, we used epidemiological and genetic data linking environmental exposure or genetic variants to phenotypic disease to construct Human Phenotype Networks and separately analyze the effects of both environment and genetic factors on disease interactions. To better capture the intricacies of the interactions between environmental exposure and the biological pathways in complex disorders, we integrate both aspects into a single “tripartite” network. Despite extensive research, the mechanisms by which chemical agents disrupt biological pathways are still poorly understood. In this study, we use our integrated network model to identify specific biological pathway candidates possibly disrupted by environmental agents. We conjecture that a higher number of co-occurrences between an environmental substance and biological pathway pair can be associated with a higher likelihood that the substance is involved in disrupting that pathway. We validate our model by demonstrating its ability to detect known arsenic and signal transduction pathway interactions and speculate on candidate cell-cell junction organization pathways disrupted by cadmium. The validation was supported by distinct publications of cell biology and genetic studies that associated environmental exposure to pathway disruption. The integrated network approach is a novel method for detecting the biological effects of environmental exposures. A better understanding of the molecular processes associated with specific environmental exposures will help in developing targeted molecular therapies for patients who have been exposed to the toxicity of environmental chemicals.

Biomedicine produces copious information it cannot fully exploit. Specifically, there is considerable need to integrate knowledge from disparate studies to discover connections across domains. Here, we used a Collaborative Filtering approach, inspired by online recommendation algorithms, in which non-negative matrix factorization (NMF) predicts interactions among chemicals, genes, and diseases only from pairwise information about their interactions. Our approach, applied to matrices derived from the Comparative Toxicogenomics Database, successfully recovered Chemical-Disease, Chemical-Gene, and Disease-Gene networks in 10-fold cross-validation experiments. Additionally, we could predict each of these interaction matrices from the other two. Integrating all three CTD interaction matrices with NMF led to good predictions of STRING, an independent, external network of protein-protein interactions. Finally, this approach could integrate the CTD and STRING interaction data to improve Chemical-Gene cross-validation performance significantly, and, in a time-stamped study, it predicted information added to CTD after a given date, using only data prior to that date. We conclude that collaborative filtering can integrate information across multiple types of biological entities, and that as a first step towards precision medicine it can compute drug repurposing hypotheses.

We have previously developed a statistical method to identify gene sets enriched with condition-specific genetic dependencies. The method constructs gene dependency networks from bootstrapped samples in one condition and computes the divergence between distributions of network likelihood scores from different conditions. It was shown to be capable of sensitive and specific identification of pathways with phenotype-specific dysregulation, i.e., rewiring of dependencies between genes in different conditions. We now present an extension of the method by incorporating prior knowledge into the inference of networks. The degree of prior knowledge incorporation has substantial effect on the sensitivity of the method, as the data is the source of condition specificity while prior knowledge incorporation can provide additional support for dependencies that are only partially supported by the data. Use of prior knowledge also significantly improved the interpretability of the results. Further analysis of topological characteristics of gene differential dependency networks provides a new approach to identify genes that could play important roles in biological signaling in a specific condition, hence, promising targets customized to a specific condition. Through analysis of TCGA glioblastoma multiforme data, we demonstrate the method can identify not only potentially promising targets but also underlying biology for new targets.

The human protein kinome presents one of the largest protein families that orchestrate functional processes in complex cellular networks, and when perturbed, can cause various cancers. The abundance and diversity of genetic, structural, and biochemical data underlies the complexity of mechanisms by which targeted and personalized drugs can combat mutational profiles in protein kinases. Coupled with the evolution of system biology approaches, genomic and proteomic technologies are rapidly identifying and charactering novel resistance mechanisms with the goal to inform rationale design of personalized kinase drugs. Integration of experimental and computational approaches can help to bring these data into a unified conceptual framework and develop robust models for predicting the clinical drug resistance. In the current study, we employ a battery of synergistic computational approaches that integrate genetic, evolutionary, biochemical, and structural data to characterize the effect of cancer mutations in protein kinases. We provide a detailed structural classification and analysis of genetic signatures associated with oncogenic mutations. By integrating genetic and structural data, we employ network modeling to dissect mechanisms of kinase drug sensitivities to oncogenic EGFR mutations. Using biophysical simulations and analysis of protein structure networks, we show that conformational-specific drug binding of Lapatinib may elicit resistant mutations in the EGFR kinase that are linked with the ligand-mediated changes in the residue interaction networks and global network properties of key residues that are responsible for structural stability of specific functional states. A strong network dependency on high centrality residues in the conformation-specific Lapatinib-EGFR complex may explain vulnerability of drug binding to a broad spectrum of mutations and the emergence of drug resistance. Our study offers a systems-based perspective on drug design by unravelling complex relationships between robustness of targeted kinase genes and binding specificity of targeted kinase drugs. We discuss how these approaches can exploit advances in chemical biology and network science to develop novel strategies for rationally tailored and robust personalized drug therapies.

Association studies have shown and continue to show a substantial amount of success in identifying links between multiple single nucleotide polymorphisms (SNPs) and phenotypes. These studies are also believed to provide insights toward identification of new drug targets and therapies. Albeit of all the success, challenges still remain for applying and prioritizing these associations based on available biological knowledge. Along with single variant association analysis, genetic interactions also play an important role in uncovering the etiology and progression of complex traits. For gene-gene interaction analysis, selection of the variants to test for associations still poses a challenge in identifying epistatic interactions among the large list of variants available in high-throughput, genome-wide datasets. Therefore in this study, we propose a pipeline to identify interactions among genetic variants that are associated with multiple phenotypes by prioritizing previously published results from main effect association analysis (genome-wide and phenome-wide association analysis) based on a-priori biological knowledge in AIDS Clinical Trials Group (ACTG) data. We approached the prioritization and filtration of variants by using the results of a previously published single variant PheWAS and then utilizing biological information from the Roadmap Epigenome project. We removed variants in low functional activity regions based on chromatin states annotation and then conducted an exhaustive pairwise interaction search using linear regression analysis. We performed this analysis in two independent pre-treatment clinical trial datasets from ACTG to allow for both discovery and replication. Using a regression framework, we observed 50,798 associations that replicate at p-value 0.01 for 26 phenotypes, among which 2,176 associations for 212 unique SNPs for fasting blood glucose phenotype reach Bonferroni significance and an additional 9,970 interactions for high-density lipoprotein (HDL) phenotype and fasting blood glucose (total of 12,146 associations) reach FDR significance. We conclude that this method of prioritizing variants to look for epistatic interactions can be used extensively for generating hypotheses for genomewide and phenome-wide interaction analyses. This original Phenome-wide Interaction study (PheWIS) can be applied further to patients enrolled in randomized clinical trials to establish the relationship between patient's response to a particular drug therapy and non-linear combination of variants that might be affecting the outcome.

Understanding community structure in networks has received considerable attention in recent years. Detecting and leveraging community structure holds promise for understanding and potentially intervening with the spread of influence. Network features of this type have important implications in a number of research areas, including, marketing, social networks, and biology. However, an overwhelming majority of traditional approaches to community detection cannot readily incorporate information of node attributes. Integrating structural and attribute information is a major challenge. We propose a exible iterative method; inverse regularized Markov Clustering (irMCL), to network clustering via the manipulation of the transition probability matrix (aka stochastic flow) corresponding to a graph. Similar to traditional Markov Clustering, irMCL iterates between “expand” and “inflate” operations, which aim to strengthen the intra-cluster flow, while weakening the inter-cluster flow. Attribute information is directly incorporated into the iterative method through a sigmoid (logistic function) that naturally dampens attribute influence that is contradictory to the stochastic flow through the network. We demonstrate advantages and the exibility of our approach using simulations and real data. We highlight an application that integrates breast cancer gene expression data set and a functional network defined via KEGG pathways reveal significant modules for survival.

Interactions between drugs, drug targets or diseases can be predicted on the basis of molecular, clinical and genomic features by, for example, exploiting similarity of disease pathways, chemical structures, activities across cell lines or clinical manifestations of diseases. A successful way to better understand complex interactions in biomedical systems is to employ collective relational learning approaches that can jointly model diverse relationships present in multiplex data. We propose a novel collective pairwise classification approach for multi-way data analysis. Our model leverages the superiority of latent factor models and classifies relationships in a large relational data domain using a pairwise ranking loss. In contrast to current approaches, our method estimates probabilities, such that probabilities for existing relationships are higher than for assumed-to-be-negative relationships. Although our method bears correspondence with the maximization of non-differentiable area under the ROC curve, we were able to design a learning algorithm that scales well on multi-relational data encoding interactions between thousands of entities.We use the new method to infer relationships from multiplex drug data and to predict connections between clinical manifestations of diseases and their underlying molecular signatures. Our method achieves promising predictive performance when compared to state-of-the-art alternative approaches and can make “category-jumping” predictions about diseases from genomic and clinical data generated far outside the molecular context.

The following sections are included:

• Introduction
• Session Summary

Previous candidate gene and genome-wide association studies have identified common genetic variants in LPA associated with the quantitative trait Lp(a), an emerging risk factor for cardiovascular disease. These associations are population-specific and many have not yet been tested for association with the clinical outcome of interest. To fill this gap in knowledge, we accessed the epidemiologic Third National Health and Nutrition Examination Surveys (NHANES III) and BioVU, the Vanderbilt University Medical Center biorepository linked to de-identified electronic health records (EHRs), including billing codes (ICD-9-CM) and clinical notes, to test population-specific Lp(a)-associated variants for an association with myocardial infarction (MI) among African Americans. We performed electronic phenotyping among African Americans in BioVU≥40 years of age using billing codes. At total of 93 cases and 522 controls were identified in NHANES III and 265 cases and 363 controls were identified in BioVU. We tested five known Lp(a)-associated genetic variants (rs1367211, rs41271028, rs6907156, rs10945682, and rs1652507) in both NHANES III and BioVU for association with myocardial infarction. We also tested LPA rs3798220 (I4399M), previously associated with increased levels of Lp(a), MI, and coronary artery disease in European Americans, in BioVU. After meta-analysis, tests of association using logistic regression assuming an additive genetic model revealed no significant associations (p<0.05) for any of the five LPA variants previously associated with Lp(a) levels in African Americans. Also, I4399M rs3798220 was not associated with MI in African Americans (odds ratio = 0.51; 95% confidence interval: 0.16 − 1.65; p=0.26) despite strong, replicated associations with MI and coronary artery disease in European American genome-wide association studies. These data highlight the challenges in translating quantitative trait associations to clinical outcomes in diverse populations using large epidemiologic and clinic-based collections as envisioned for the Precision Medicine Initiative.

Many recent imaging genetic studies focus on detecting the associations between genetic markers such as single nucleotide polymorphisms (SNPs) and quantitative traits (QTs). Although there exist a large number of generalized multivariate regression analysis methods, few of them have used diagnosis information in subjects to enhance the analysis performance. In addition, few of models have investigated the identification of multi-modality phenotypic patterns associated with interesting genotype groups in traditional methods. To reveal disease-relevant imaging genetic associations, we propose a novel diagnosis-guided multi-modality (DGMM) framework to discover multi-modality imaging QTs that are associated with both Alzheimer's disease (AD) and its top genetic risk factor (i.e., APOE SNP rs429358). The strength of our proposed method is that it explicitly models the priori diagnosis information among subjects in the objective function for selecting the disease-relevant and robust multi-modality QTs associated with the SNP. We evaluate our method on two modalities of imaging phenotypes, i.e., those extracted from structural magnetic resonance imaging (MRI) data and fluorodeoxyglucose positron emission tomography (FDG-PET) data in the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. The experimental results demonstrate that our proposed method not only achieves better performances under the metrics of root mean squared error and correlation coefficient but also can identify common informative regions of interests (ROIs) across multiple modalities to guide the disease-induced biological interpretation, compared with other reference methods.

Osteoarthritis (OA) significantly compromises the life quality of affected individuals and imposes a substantial economic burden on our society. Unfortunately the pathogenesis of the disease is till poorly understood and no effective medications have been developed. OA is a complex disease that involves both genetic and environmental influences. To elucidate the complex interlinked structure of metabolic processes associated with OA, we developed a differential correlation network approach to detecting the interconnection of metabolite pairs whose relationships are significantly altered due to the diseased process. Through topological analysis of such a differential network, we identified key metabolites that played an important role in governing the connectivity and information flow of the network. Identification of these key metabolites suggests the association of their underlying cellular processes with OA and may help elucidate the pathogenesis of the disease and the development of novel targeted therapies.

There is growing use of ontologies for the measurement of cross-species phenotype similarity. Such similarity measurements contribute to diverse applications, such as identifying genetic models for human diseases, transferring knowledge among model organisms, and studying the genetic basis of evolutionary innovations. Two organismal features, whether genes, anatomical parts, or any other inherited feature, are considered to be homologous when they are evolutionarily derived from a single feature in a common ancestor. A classic example is the homology between the paired fins of fishes and vertebrate limbs. Anatomical ontologies that model the structural relations among parts may fail to include some known anatomical homologies unless they are deliberately added as separate axioms. The consequences of neglecting known homologies for applications that rely on such ontologies has not been well studied. Here, we examine how semantic similarity is affected when external homology knowledge is included. We measure phenotypic similarity between orthologous and non-orthologous gene pairs between humans and either mouse or zebrafish, and compare the inclusion of real with faux homology axioms. Semantic similarity was preferentially increased for orthologs when using real homology axioms, but only in the more divergent of the two species comparisons (human to zebrafish, not human to mouse), and the relative increase was less than 1% to non-orthologs. By contrast, inclusion of both real and faux random homology axioms preferentially increased similarities between genes that were initially more dissimilar in the other comparisons. Biologically meaningful increases in semantic similarity were seen for a select subset of gene pairs. Overall, the effect of including homology axioms on cross-species semantic similarity was modest at the levels of divergence examined here, but our results hint that it may be greater for more distant species comparisons.

The practice of medicine is predicated on discovering commonalities or distinguishing characteristics among patients to inform corresponding treatment. Given a patient grouping (hereafter referred to as a phenotype), clinicians can implement a treatment pathway accounting for the underlying cause of disease in that phenotype. Traditionally, phenotypes have been discovered by intuition, experience in practice, and advancements in basic science, but these approaches are often heuristic, labor intensive, and can take decades to produce actionable knowledge. Although our understanding of disease has progressed substantially in the past century, there are still important domains in which our phenotypes are murky, such as in behavioral health or in hospital settings. To accelerate phenotype discovery, researchers have used machine learning to find patterns in electronic health records, but have often been thwarted by missing data, sparsity, and data heterogeneity. In this study, we use a flexible framework called Generalized Low Rank Modeling (GLRM) to overcome these barriers and discover phenotypes in two sources of patient data. First, we analyze data from the 2010 Healthcare Cost and Utilization Project National Inpatient Sample (NIS), which contains upwards of 8 million hospitalization records consisting of administrative codes and demographic information. Second, we analyze a small (N=1746), local dataset documenting the clinical progression of autism spectrum disorder patients using granular features from the electronic health record, including text from physician notes. We demonstrate that low rank modeling successfully captures known and putative phenotypes in these vastly different datasets.

We present a computational strategy to simulate drug treatment in a personalized setting. The method is based on integrating patient mutation and differential expression data with a protein-protein interaction network. We test the impact of in-silico deletions of different proteins on the flow of information in the network and use the results to infer potential drug targets. We apply our method to AML data from TCGA and validate the predicted drug targets using known targets. To benchmark our patient-specific approach, we compare the personalized setting predictions to those of the conventional setting. Our predicted drug targets are highly enriched with known targets from DrugBank and COSMIC (p < 10⁻⁵ outperforming the non-personalized predictions. Finally, we focus on the largest AML patient subgroup (~30%) which is characterized by an FLT3 mutation, and utilize our prediction score to rank patient sensitivity to inhibition of each predicted target, reproducing previous findings of in-vitro experiments.

Electronic health records (EHR) provide a comprehensive resource for discovery, allowing unprecedented exploration of the impact of genetic architecture on health and disease. The data of EHRs also allow for exploration of the complex interactions between health measures across health and disease. The discoveries arising from EHR based research provide important information for the identification of genetic variation for clinical decision-making. Due to the breadth of information collected within the EHR, a challenge for discovery using EHR based data is the development of high-throughput tools that expose important areas of further research, from genetic variants to phenotypes. Phenome-Wide Association studies (PheWAS) provide a way to explore the association between genetic variants and comprehensive phenotypic measurements, generating new hypotheses and also exposing the complex relationships between genetic architecture and outcomes, including pleiotropy. EHR based PheWAS have mainly evaluated associations with case/control status from International Classification of Disease, Ninth Edition (ICD-9) codes. While these studies have highlighted discovery through PheWAS, the rich resource of clinical lab measures collected within the EHR can be better utilized for highthroughput PheWAS analyses and discovery. To better use these resources and enrich PheWAS association results we have developed a sound methodology for extracting a wide range of clinical lab measures from EHR data. We have extracted a first set of 21 clinical lab measures from the de-identified EHR of participants of the Geisinger MyCode^TM biorepository, and calculated the median of these lab measures for 12,039 subjects. Next we evaluated the association between these 21 clinical lab median values and 635,525 genetic variants, performing a genome-wide association study (GWAS) for each of 21 clinical lab measures. We then calculated the association between SNPs from these GWAS passing our Bonferroni defined p-value cutoff and 165 ICD-9 codes. Through the GWAS we found a series of results replicating known associations, and also some potentially novel associations with less studied clinical lab measures. We found the majority of the PheWAS ICD-9 diagnoses highly related to the clinical lab measures associated with same SNPs. Moving forward, we will be evaluating further phenotypes and expanding the methodology for successful extraction of clinical lab measurements for research and PheWAS use. These developments are important for expanding the PheWAS approach for improved EHR based discovery.

During January 2015, President Obama announced the Precision Medicine Initiative [1], strengthening communal efforts to integrate patient-centric molecular, environmental, and clinical “big” data. Such efforts have already improved aspects of clinical management for diseases such as non-small cell lung carcinoma [2], breast cancer [3], and hypertrophic cardiomyopathy [4]. To maintain this track record, it is necessary to cultivate practices that ensure reproducibility as large-scale heterogeneous datasets and databases proliferate. For example, the NIH has outlined initiatives to enhance reproducibility in preclinical research [5], both Science [6] and Nature [7] have featured recent editorials on reproducibility, and several authors have noted the issues of utilizing big data for public health [8], but few methods exist to ensure that big data resources motivated by precision medicine are being used reproducibly. Relevant challenges include: (1) integrative analyses of heterogeneous measurement platforms (e.g. genomic, clinical, quantified self, and exposure data), (2) the tradeoff in making personalized decisions using more targeted (e.g. individual-level) but potentially much noisier subsets of data, and (3) the unprecedented scale of asynchronous observational and population level inquiry (i.e. many investigators separately mining shared/publicly-available data)…

This article presents a reproducible research workflow for amplicon-based microbiome studies in personalized medicine created using Bioconductor packages and the knitr markdown interface.We show that sometimes a multiplicity of choices and lack of consistent documentation at each stage of the sequential processing pipeline used for the analysis of microbiome data can lead to spurious results. We propose its replacement with reproducible and documented analysis using R packages dada2, knitr, and phyloseq. This workflow implements both key stages of amplicon analysis: the initial filtering and denoising steps needed to construct taxonomic feature tables from error-containing sequencing reads (dada2), and the exploratory and inferential analysis of those feature tables and associated sample metadata (phyloseq). This workow facilitates reproducible interrogation of the full set of choices required in microbiome studies. We present several examples in which we leverage existing packages for analysis in a way that allows easy sharing and modification by others, and give pointers to articles that depend on this reproducible workflow for the study of longitudinal and spatial series analyses of the vaginal microbiome in pregnancy and the oral microbiome in humans with healthy dentition and intra-oral tissues.

Automatically data-mining clinical practice patterns from electronic health records (EHR) can enable prediction of future practices as a form of clinical decision support (CDS). Our objective is to determine the stability of learned clinical practice patterns over time and what implication this has when using varying longitudinal historical data sources towards predicting future decisions. We trained an association rule engine for clinical orders (e.g., labs, imaging, medications) using structured inpatient data from a tertiary academic hospital. Comparing top order associations per admission diagnosis from training data in 2009 vs. 2012, we find practice variability from unstable diagnoses with rank biased overlap (RBO)<0.35 (e.g., pneumonia) to stable admissions for planned procedures (e.g., chemotherapy, surgery) with comparatively high RBO>0.6. Predicting admission orders for future (2013) patients with associations trained on recent (2012) vs. older (2009) data improved accuracy evaluated by area under the receiver operating characteristic curve (ROC-AUC) 0.89 to 0.92, precision at ten (positive predictive value of the top ten predictions against actual orders) 30% to 37%, and weighted recall (sensitivity) at ten 2.4% to 13%, (P<10⁻¹⁰). Training with more longitudinal data (2009-2012) was no better than only using recent (2012) data. Secular trends in practice patterns likely explain why smaller but more recent training data is more accurate at predicting future practices.

When seeking to reproduce results derived from whole-exome or genome sequencing data that could advance precision medicine, the time and expense required to produce a patient cohort make data repurposing an attractive option. The first step in repurposing is setting some quality baseline for the data so that conclusions are not spurious. This is difficult because there can be variations in quality from center to center, clinic to clinic and even patient to patient. Here, we assessed the quality of the whole-exome germline mutations of TCGA cancer patients using patterns of nucleotide substitution and negative selection against impactful mutations. We estimated the fraction of false positive variant calls for each exome with respect to two gold standard germline exomes, and found large variability in the quality of SNV calls between samples, cancer subtypes, and institutions. We then demonstrated how variant features, such as the average base quality for reads supporting an allele, can be used to identify sample-specific filtering parameters to optimize the removal of false positive calls. We concluded that while these germlines have many potential applications to precision medicine, users should assess the quality of the available exome data prior to use and perform additional filtering steps.

Precision medicine requires precise evidence-based practice and precise definition of the patients included in clinical studies for evidence generalization. Clinical research exclusion criteria define confounder patient characteristics for exclusion from a study. However, unnecessary exclusion criteria can weaken patient representativeness of study designs and generalizability of study results. This paper presents a method for identifying questionable exclusion criteria for 38 mental disorders. We extracted common eligibility features (CEFs) from all trials on these disorders from ClinicalTrials.gov. Network Analysis showed scale-free property of the CEF network, indicating uneven usage frequencies among CEFs. By comparing these CEFs' term frequencies in clinical trials' exclusion criteria and in the PubMed Medical Encyclopedia for matching conditions, we identified unjustified potential overuse of exclusion CEFs in mental disorder trials. Then we discussed the limitations in current exclusion criteria designs and made recommendations for achieving more patient-centered exclusion criteria definitions.

There are now hundreds of thousands of pathogenicity assertions that relate genetic variation to disease, but most of this clinically utilized variation has no accepted quantitative disease risk estimate. Recent disease-specific studies have used control sequence data to reclassify large amounts of prior pathogenic variation, but there is a critical need to scale up both the pace and feasibility of such pathogenicity reassessments across human disease. In this manuscript we develop a shareable computational framework to quantify pathogenicity assertions. We release a reproducible “digital notebook” that integrates executable code, text annotations, and mathematical expressions in a freely accessible statistical environment. We extend previous disease-specific pathogenicity assessments to over 6,000 diseases and 160,000 assertions in the ClinVar database. Investigators can use this platform to prioritize variants for reassessment and tailor genetic model parameters (such as prevalence and heterogeneity) to expose the uncertainty underlying pathogenicity-based risk assessments. Finally, we release a website that links users to pathogenic variation for a queried disease, supporting literature, and implied disease risk calculations subject to user-defined and disease-specific genetic risk models in order to facilitate variant reassessments.

Rapid advances in personal, cohort, and population-scale data acquisition, such as via sequencing, proteomics, mass spectroscopy, biosensors, mobile health devices and social network activity and other apps are opening up new vistas for personalized health biomedical data collection, analysis and insight. To achieve the vaunted goals of precision medicine and go from measurement to clinical translation, substantial gains still need to be made in methods of data and knowledge integration, analysis, discovery and interpretation. In this session of the 2016 Pacific Symposium on Biocomputing, we present sixteen papers to help accomplish this for precision medicine.

Next-generation sequencing technology has presented an opportunity for rare variant discovery and association of these variants with disease. To address the challenges of rare variant analysis, multiple statistical methods have been developed for combining rare variants to increase statistical power for detecting associations. BioBin is an automated tool that expands on collapsing/binning methods by performing multi-level variant aggregation with a flexible, biologically informed binning strategy using an internal biorepository, the Library of Knowledge (LOKI). The databases within LOKI provide variant details, regional annotations and pathway interactions which can be used to generate bins of biologically-related variants, thereby increasing the power of any subsequent statistical test. In this study, we expand the framework of BioBin to incorporate statistical tests, including a dispersion-based test, SKAT, thereby providing the option of performing a unified collapsing and statistical rare variant analysis in one tool. Extensive simulation studies performed on gene-coding regions showed a Bin-KAT analysis to have greater power than BioBin-regression in all simulated conditions, including variants influencing the phenotype in the same direction, a scenario where burden tests often retain greater power. The use of Madsen- Browning variant weighting increased power in the burden analysis to that equitable with Bin-KAT; but overall Bin-KAT retained equivalent or higher power under all conditions. Bin-KAT was applied to a study of 82 pharmacogenes sequenced in the Marshfield Personalized Medicine Research Project (PMRP). We looked for association of these genes with 9 different phenotypes extracted from the electronic health record. This study demonstrates that Bin-KAT is a powerful tool for the identification of genes harboring low frequency variants for complex phenotypes.

Machine learning applications in precision medicine are severely limited by the scarcity of data to learn from. Indeed, training data often contains many more features than samples. To alleviate the resulting statistical issues, the multitask learning framework proposes to learn different but related tasks jointly, rather than independently, by sharing information between these tasks. Within this framework, the joint regularization of model parameters results in models with few non-zero coefficients and that share similar sparsity patterns. We propose a new regularized multitask approach that incorporates task descriptors, hence modulating the amount of information shared between tasks according to their similarity. We show on simulated data that this method outperforms other multitask feature selection approaches, particularly in the case of scarce data. In addition, we demonstrate on peptide MHC-I binding data the ability of the proposed approach to make predictions for new tasks for which no training data is available.

We propose hypothesis tests for detecting dopaminergic medication response in Parkinson disease patients, using longitudinal sensor data collected by smartphones. The processed data is composed of multiple features extracted from active tapping tasks performed by the participant on a daily basis, before and after medication, over several months. Each extracted feature corresponds to a time series of measurements annotated according to whether the measurement was taken before or after the patient has taken his/her medication. Even though the data is longitudinal in nature, we show that simple hypothesis tests for detecting medication response, which ignore the serial correlation structure of the data, are still statistically valid, showing type I error rates at the nominal level. We propose two distinct personalized testing approaches. In the first, we combine multiple feature-specific tests into a single union-intersection test. In the second, we construct personalized classifiers of the before/after medication labels using all the extracted features of a given participant, and test the null hypothesis that the area under the receiver operating characteristic curve of the classifier is equal to 1/2. We compare the statistical power of the personalized classifier tests and personalized union-intersection tests in a simulation study, and illustrate the performance of the proposed tests using data from mPower Parkinsons disease study, recently launched as part of Apples ResearchKit mobile platform. Our results suggest that the personalized tests, which ignore the longitudinal aspect of the data, can perform well in real data analyses, suggesting they might be used as a sound baseline approach, to which more sophisticated methods can be compared to.

Kidney disease is a well-known health disparity in the United States where African Americans are affected at higher rates compared with other groups such as European Americans and Mexican Americans. Common genetic variants in the myosin, heavy chain 9, non-muscle (MYH9) gene were initially identified as associated with non-diabetic end-stage renal disease in African Americans, and it is now understood that these variants are in strong linkage disequilibrium with likely causal variants in neighboring APOL1. Subsequent genome-wide and candidate gene studies have suggested that MYH9 common variants among others are also associated with chronic kidney disease and quantitative measures of kidney function in various populations. In a precision medicine setting, it is important to consider genetic effects or genetic associations that differ across racial/ethnic groups in delivering data relevant to disease risk or individual-level patient assessment. Kidney disease and quantitative trait-associated genetic variants have yet to be systematically characterized in multiple racial/ethnic groups. Therefore, to further characterize the prevalence of these genetic variants and their association with kidney related traits, we have genotyped 10 kidney disease or quantitative trait-associated single nucleotide polymorphisms (SNPs) (rs2900976, rs10505955, rs10502868, rs1243400, rs9305354, rs12917707, rs17319721, rs2467853, rs2032487, and rs4821480) in 14,998 participants from the population-based cross-sectional National Health and Nutrition Examination Surveys (NHANES) III and 1999-2002 as part of the Epidemiologic Architecture for Genes Linked to Environment (EAGLE) study. In this general adult population ascertained regardless of health status (6,293 non-Hispanic whites, 3,013 non-Hispanic blacks, and 3,542 Mexican Americans), we observed higher rates of chronic kidney disease among non-Hispanic blacks compared with the other groups as expected. We performed single SNP tests of association using linear regressions assuming an additive genetic model adjusted for age, sex, diastolic blood pressure, systolic blood pressure, and type 2 diabetes status for several outcomes including creatinine (urinary), creatinine (serum), albumin (urinary), eGFR, and albumin-to-urinary creatinine ratio (ACR). We also tested for associations between each SNP and chronic kidney disease and albuminuria using logistic regression. Surprisingly, none of the MYH9 variants tested was associated with kidney diseases or traits in non-Hispanic blacks (p>0.05), perhaps attributable to the clinical heterogeneity of kidney disease in this population. Several associations were observed in each racial/ethnic group at p<0.05, but none were consistently associated in the same direction in all three groups. The lack of significant and consistent associations is most likely due to power highlighting the importance of the availability of large, diverse populations for genetic association studies of complex diseases and traits to inform precision medicine efforts in diverse patient populations.

Glial tumors have been heavily studied and sequenced, leading to scores of findings about altered genes. This explosion in knowledge has not been matched with clinical success, but efforts to understand the synergies between drivers of glial tumors may alleviate the situation. We present a novel molecular classification system that captures the combinatorial nature of relationships between alterations in these diseases. We use this classification to mine for enrichment of variants of unknown significance, and demonstrate a method for segregating unknown variants with functional importance from passengers and SNPs.

Open clinical trial data offer many opportunities for the scientific community to independently verify published results, evaluate new hypotheses and conduct meta-analyses. These data provide a springboard for scientific advances in precision medicine but the question arises as to how representative clinical trials data are of cancer patients overall. Here we present the integrative analysis of data from several cancer clinical trials and compare these to patient-level data from The Cancer Genome Atlas (TCGA). Comparison of cancer type-specific survival rates reveals that these are overall lower in trial subjects. This effect, at least to some extent, can be explained by the more advanced stages of cancer of trial subjects. This analysis also reveals that for stage IV cancer, colorectal cancer patients have a better chance of survival than breast cancer patients. On the other hand, for all other stages, breast cancer patients have better survival than colorectal cancer patients. Comparison of survival in different stages of disease between the two datasets reveals that subjects with stage IV cancer from the trials dataset have a lower chance of survival than matching stage IV subjects from TCGA. One likely explanation for this observation is that stage IV trial subjects have lower survival rates since their cancer is less likely to respond to treatment. To conclude, we present here a newly available clinical trials dataset which allowed for the integration of patient-level data from many cancer clinical trials. Our comprehensive analysis reveals that cancer-related clinical trials are not representative of general cancer patient populations, mostly due to their focus on the more advanced stages of the disease. These and other limitations of clinical trials data should, perhaps, be taken into consideration in medical research and in the field of precision medicine.

According to Cancer Research UK, cancer is a leading cause of death accounting for more than one in four of all deaths in 2011. The recent advances in experimental technologies in cancer research have resulted in the accumulation of large amounts of patient-specific datasets, which provide complementary information on the same cancer type. We introduce a versatile data fusion (integration) framework that can effectively integrate somatic mutation data, molecular interactions and drug chemical data to address three key challenges in cancer research: stratification of patients into groups having different clinical outcomes, prediction of driver genes whose mutations trigger the onset and development of cancers, and repurposing of drugs treating particular cancer patient groups. Our new framework is based on graph-regularised non-negative matrix tri-factorization, a machine learning technique for co-clustering heterogeneous datasets. We apply our framework on ovarian cancer data to simultaneously cluster patients, genes and drugs by utilising all datasets.We demonstrate superior performance of our method over the state-of-the-art method, Network-based Stratification, in identifying three patient subgroups that have significant differences in survival outcomes and that are in good agreement with other clinical data. Also, we identify potential new driver genes that we obtain by analysing the gene clusters enriched in known drivers of ovarian cancer progression. We validated the top scoring genes identified as new drivers through database search and biomedical literature curation. Finally, we identify potential candidate drugs for repurposing that could be used in treatment of the identified patient subgroups by targeting their mutated gene products. We validated a large percentage of our drug-target predictions by using other databases and through literature curation.

Neuropsychiatric disorders are the leading cause of disability worldwide and there is no gold standard currently available for the measurement of mental health. This issue is exacerbated by the fact that the information physicians use to diagnose these disorders is episodic and often subjective. Current methods to monitor mental health involve the use of subjective DSM-5 guidelines, and advances in EEG and video monitoring technologies have not been widely adopted due to invasiveness and inconvenience. Wearable technologies have surfaced as a ubiquitous and unobtrusive method for providing continuous, quantitative data about a patient. Here, we introduce PRISM—Passive, Real-time Information for Sensing Mental Health. This platform integrates motion, light and heart rate data from a smart watch application with user interactions and text entries from a web application. We have demonstrated a proof of concept by collecting preliminary data through a pilot study of 13 subjects. We have engineered appropriate features and applied both unsupervised and supervised learning to develop models that are predictive of user-reported ratings of their emotional state, demonstrating that the data has the potential to be useful for evaluating mental health. This platform could allow patients and clinicians to leverage continuous streams of passive data for early and accurate diagnosis as well as constant monitoring of patients suffering from mental disorders.

The move from Empirical Medicine towards Personalized Medicine has attracted attention to Stratified Medicine (SM). Some methods are provided in the literature for patient stratification, which is the central task of SM, however, there are still significant open issues. First, it is still unclear if integrating different datatypes will help in detecting disease subtypes more accurately, and, if not, which datatype(s) are most useful for this task. Second, it is not clear how we can compare different methods of patient stratification. Third, as most of the proposed stratification methods are deterministic, there is a need for investigating the potential benefits of applying probabilistic methods. To address these issues, we introduce a novel integrative Bayesian biclustering method, called B2PS, for patient stratification and propose methods for evaluating the results. Our experimental results demonstrate the superiority of B2PS over a popular state-of-the-art method and the benefits of Bayesian approaches. Our results agree with the intuition that transcriptomic data forms a better basis for patient stratification than genomic data.

Recent studies on copy number variation (CNV) have suggested that an increasing burden of CNVs is associated with susceptibility or resistance to disease. A large number of genes or genomic loci contribute to complex diseases such as autism. Thus, total genomic copy number burden, as an accumulation of copy number change, is a meaningful measure of genomic instability to identify the association between global genetic effects and phenotypes of interest. However, no systematic annotation pipeline has been developed to interpret biological meaning based on the accumulation of copy number change across the genome associated with a phenotype of interest. In this study, we develop a comprehensive and systematic pipeline for annotating copy number variants into genes/genomic regions and subsequently pathways and other gene groups using Biofilter – a bioinformatics tool that aggregates over a dozen publicly available databases of prior biological knowledge. Next we conduct enrichment tests of biologically defined groupings of CNVs including genes, pathways, Gene Ontology, or protein families. We applied the proposed pipeline to a CNV dataset from the Marshfield Clinic Personalized Medicine Research Project (PMRP) in a quantitative trait phenotype derived from the electronic health record – total cholesterol. We identified several significant pathways such as toll-like receptor signaling pathway and hepatitis C pathway, gene ontologies (GOs) of nucleoside triphosphatase activity (NTPase) and response to virus, and protein families such as cell morphogenesis that are associated with the total cholesterol phenotype based on CNV profiles (permutation p-value < 0.01). Based on the copy number burden analysis, it follows that the more and larger the copy number changes, the more likely that one or more target genes that influence disease risk and phenotypic severity will be affected. Thus, our study suggests the proposed enrichment pipeline could improve the interpretability of copy number burden analysis where hundreds of loci or genes contribute toward disease susceptibility via biological knowledge groups such as pathways. This CNV annotation pipeline with Biofilter can be used for CNV data from any genotyping or sequencing platform and to explore CNV enrichment for any traits or phenotypes. Biofilter continues to be a powerful bioinformatics tool for annotating, filtering, and constructing biologically informed models for association analysis – now including copy number variants.

Access and utilization of electronic health records with extensive medication lists and genetic profiles is rapidly advancing discoveries in pharmacogenomics. In this study, we analyzed ~116,000 variants on the Illumina Metabochip for response to antihypertensive and lipid lowering medications in African American adults from BioVU, the Vanderbilt University Medical Center's biorepository linked to de-identified electronic health records. Our study population included individuals who were prescribed an antihypertensive or lipid lowering medication, and who had both pre- and post-medication blood pressure or low-density lipoprotein cholesterol (LDL-C) measurements, respectively. Among those with pre- and post-medication systolic and diastolic blood pressure measurements (n=2,268), the average change in systolic and diastolic blood pressure was −0.6 mg Hg and −0.8 mm Hg, respectively. Among those with pre- and post-medication LDL-C measurements (n=1,244), the average change in LDL-C was −26.3 mg/dL. SNPs were tested for an association with change and percent change in blood pressure or blood levels of LDL-C. After adjustment for multiple testing, we did not observe any significant associations, and we were not able to replicate previously reported associations, such as in APOE and LPA, from the literature. The present study illustrates the benefits and challenges with using electronic health records linked to biorepositories for pharmacogenomic studies.

The causes of complex diseases are multifactorial and the phenotypes of complex diseases are typically heterogeneous, posting significant challenges for both the experiment design and statistical inference in the study of such diseases. Transcriptome profiling can potentially provide key insights on the pathogenesis of diseases, but the signals from the disease causes and consequences are intertwined, leaving it to speculations what are likely causal. Genome-wide association study on the other hand provides direct evidences on the potential genetic causes of diseases, but it does not provide a comprehensive view of disease pathogenesis, and it has difficulties in detecting the weak signals from individual genes. Here we propose an approach diseaseExPatho that combines transcriptome data, regulome knowledge, and GWAS results if available, for separating the causes and consequences in the disease transcriptome. DiseaseExPatho computationally deconvolutes the expression data into gene expression modules, hierarchically ranks the modules based on regulome using a novel algorithm, and given GWAS data, it directly labels the potential causal gene modules based on their correlations with genome-wide gene-disease associations. Strikingly, we observed that the putative causal modules are not necessarily differentially expressed in disease, while the other modules can show strong differential expression without enrichment of top GWAS variations. On the other hand, we showed that the regulatory network based module ranking prioritized the putative causal modules consistently in 6 diseases. We suggest that the approach is applicable to other common and rare complex diseases to prioritize causal pathways with or without genome-wide association studies.

We present a feature allocation model to reconstruct tumor subclones based on mutation pairs. The key innovation lies in the use of a pair of proximal single nucleotide variants (SNVs) for the subclone reconstruction as opposed to a single SNV. Using the categorical extension of the Indian buffet process (cIBP) we define the subclones as a vector of categorical matrices corresponding to a set of mutation pairs. Through Bayesian inference we report posterior probabilities of the number, genotypes and population frequencies of subclones in one or more tumor sample. We demonstrate the proposed methods using simulated and real-world data. A free software package is available at http://www.compgenome.org/pairclone.

The cellular composition of a tumor greatly influences the growth, spread, immune activity, drug response, and other aspects of the disease. Tumor cells are usually comprised of a heterogeneous mixture of subclones, each of which could contain their own distinct character. The presence of minor subclones poses a serious health risk for patients as any one of them could harbor a fitness advantage with respect to the current treatment regimen, fueling resistance. It is therefore vital to accurately assess the make-up of cell states within a tumor biopsy. Transcriptome-wide assays from RNA sequencing provide key data from which cell state signatures can be detected. However, the challenge is to find them within samples containing mixtures of cell types of unknown proportions. We propose a novel one-class method based on logistic regression and show that its performance is competitive to two established SVM-based methods for this detection task. We demonstrate that one-class models are able to identify specific cell types in heterogeneous cell populations better than their binary predictor counterparts. We derive one-class predictors for the major breast and bladder subtypes and reaffirm the connection between these two tissues. In addition, we use a one-class predictor to quantitatively associate an embryonic stem cell signature with an aggressive breast cancer subtype that reveals shared stemness pathways potentially important for treatment.

Realization of precision medicine ideas requires significant research effort to be able to spot subtle differences in complex diseases at the molecular level to develop personalized therapies. It is especially important in many cases of highly heterogeneous cancers. Precision diagnostics and therapeutics of such diseases demands interrogation of vast amounts of biological knowledge coupled with novel analytic methodologies. For instance, pathway-based approaches can shed light on the way tumorigenesis takes place in individual patient cases and pinpoint to novel drug targets. However, comprehensive analysis of hundreds of pathways and thousands of genes creates a combinatorial explosion, that is challenging for medical practitioners to handle at the point of care. Here we extend our previous work on mapping clinical omics data to curated Resource Description Framework (RDF) knowledge bases to derive influence diagrams of interrelationships of biomarker proteins, diseases and signal transduction pathways for personalized theranostics. We present RDF Sketch Maps – a computational method to reduce knowledge complexity for precision medicine analytics. The method of RDF Sketch Maps is inspired by the way a sketch artist conveys only important visual information and discards other unnecessary details. In our case, we compute and retain only so-called RDF Edges – places with highly important diagnostic and therapeutic information. To do this we utilize 35 maps of human signal transduction pathways by transforming 300 KEGG maps into highly processable RDF knowledge base. We have demonstrated potential clinical utility of RDF Sketch Maps in hematopoietic cancers, including analysis of pathways associated with Hairy Cell Leukemia (HCL) and Chronic Myeloid Leukemia (CML) where we achieved up to 20-fold reduction in the number of biological entities to be analyzed, while retaining most likely important entities. In experiments with pathways associated with HCL a generated RDF Sketch Map of the top 30% paths retained important information about signaling cascades leading to activation of proto-oncogene BRAF, which is usually associated with a different cancer, melanoma. Recent reports of successful treatments of HCL patients by the BRAF-targeted drug vemurafenib support the validity of the RDF Sketch Maps findings. We therefore believe that RDF Sketch Maps will be invaluable for hypothesis generation for precision diagnostics and therapeutics as well as drug repurposing studies.

Advances in both experimental and computational approaches to genome-wide analysis of RNA transcripts have dramatically expanded our understanding of the ubiquitous and diverse roles of regulatory non-coding RNAs. This conference session includes presentations exploring computational approaches for detecting regulatory RNAs in RNA-Seq data, for analyzing in vivo CLIP data on RNA-protein interactions, and for predicting interfacial residues involved in RNA-protein recognition in RNA–protein complexes and interaction networks.

CLIP-Seq protocols such as PAR-CLIP, HITS-CLIP or iCLIP allow a genome-wide analysis of protein-RNA interactions. For the processing of the resulting short read data, various tools are utilized. Some of these tools were specifically developed for CLIP-Seq data, whereas others were designed for the analysis of RNA-Seq data. To this date, however, it has not been assessed which of the available tools are most appropriate for the analysis of CLIP-Seq data. This is because an experimental gold standard dataset on which methods can be accessed and compared, is still not available. To address this lack of a gold-standard dataset, we here present Cseq-Simulator, a simulator for PAR-CLIP, HITS-CLIP and iCLIP-data. This simulator can be applied to generate realistic datasets that can serve as surrogates for experimental gold standard dataset. In this work, we also show how Cseq-Simulator can be used to perform a comparison of steps of typical CLIP-Seq analysis pipelines, such as the read alignment or the peak calling. These comparisons show which tools are useful in different settings and also allow identifying pitfalls in the data analysis.

Efforts to predict interfacial residues in protein-RNA complexes have largely focused on predicting RNA-binding residues in proteins. Computational methods for predicting protein-binding residues in RNA sequences, however, are a problem that has received relatively little attention to date. Although the value of sequence motifs for classifying and annotating protein sequences is well established, sequence motifs have not been widely applied to predicting interfacial residues in macromolecular complexes. Here, we propose a novel sequence motif-based method for “partner-specific” interfacial residue prediction. Given a specific protein-RNA pair, the goal is to simultaneously predict RNA binding residues in the protein sequence and protein-binding residues in the RNA sequence. In 5-fold cross validation experiments, our method, PS-PRIP, achieved 92% Specificity and 61% Sensitivity, with a Matthews correlation coefficient (MCC) of 0.58 in predicting RNA-binding sites in proteins. The method achieved 69% Specificity and 75% Sensitivity, but with a low MCC of 0.13 in predicting protein binding sites in RNAs. Similar performance results were obtained when PS-PRIP was tested on two independent “blind” datasets of experimentally validated protein- RNA interactions, suggesting the method should be widely applicable and valuable for identifying potential interfacial residues in protein-RNA complexes for which structural information is not available. The PS-PRIP webserver and datasets are available at: http://pridb.gdcb.iastate.edu/PSPRIP/.

Small non-coding RNAs (sRNAs) are regulatory RNA molecules that have been identified in a multitude of bacterial species and shown to control numerous cellular processes through various regulatory mechanisms. In the last decade, next generation RNA sequencing (RNA-seq) has been used for the genome-wide detection of bacterial sRNAs. Here we describe sRNA-Detect, a novel approach to identify expressed small transcripts from prokaryotic RNA-seq data. Using RNA-seq data from three bacterial species and two sequencing platforms, we performed a comparative assessment of five computational approaches for the detection of small transcripts. We demonstrate that sRNA-Detect improves upon current standalone computational approaches for identifying novel small transcripts in bacteria.

This paper describes topics pertaining to the session, “Social Media Mining for Public Health Monitoring and Surveillance,” at the Pacific Symposium on Biocomputing (PSB) 2016. In addition to summarizing the content of the session, this paper also surveys recent research on using social media data to study public health. The survey is organized into sections describing recent progress in public health problems, computational methods, and social implications.

Rapid increases in e-cigarette use and potential exposure to harmful byproducts have shifted public health focus to e-cigarettes as a possible drug of abuse. Effective surveillance of use and prevalence would allow appropriate regulatory responses. An ideal surveillance system would collect usage data in real time, focus on populations of interest, include populations unable to take the survey, allow a breadth of questions to answer, and enable geo-location analysis. Social media streams may provide this ideal system. To realize this use case, a foundational question is whether we can detect e-cigarette use at all. This work reports two pilot tasks using text classification to identify automatically Tweets that indicate e-cigarette use and/or e-cigarette use for smoking cessation. We build and define both datasets and compare performance of 4 state of the art classifiers and a keyword search for each task. Our results demonstrate excellent classifier performance of up to 0.90 and 0.94 area under the curve in each category. These promising initial results form the foundation for further studies to realize the ideal surveillance solution.

Much recent research aims to identify evidence for Drug-Drug Interactions (DDI) and Adverse Drug reactions (ADR) from the biomedical scientific literature. In addition to this "Bibliome", the universe of social media provides a very promising source of large-scale data that can help identify DDI and ADR in ways that have not been hitherto possible. Given the large number of users, analysis of social media data may be useful to identify under-reported, population-level pathology associated with DDI, thus further contributing to improvements in population health. Moreover, tapping into this data allows us to infer drug interactions with natural products—including cannabis—which constitute an array of DDI very poorly explored by biomedical research thus far.

Our goal is to determine the potential of Instagram for public health monitoring and surveillance for DDI, ADR, and behavioral pathology at large. Most social media analysis focuses on Twitter and Facebook, but Instagram is an increasingly important platform, especially among teens, with unrestricted access of public posts, high availability of posts with geolocation coordinates, and images to supplement textual analysis.

Using drug, symptom, and natural product dictionaries for identification of the various types of DDI and ADR evidence, we have collected close to 7000 user timelines spanning from October 2010 to June 2015.We report on 1) the development of a monitoring tool to easily observe user-level timelines associated with drug and symptom terms of interest, and 2) population-level behavior via the analysis of co-occurrence networks computed from user timelines at three different scales: monthly, weekly, and daily occurrences. Analysis of these networks further reveals 3) drug and symptom direct and indirect associations with greater support in user timelines, as well as 4) clusters of symptoms and drugs revealed by the collective behavior of the observed population.

This demonstrates that Instagram contains much drug- and pathology specific data for public health monitoring of DDI and ADR, and that complex network analysis provides an important toolbox to extract health-related associations and their support from large-scale social media data.

Online social media microblogs may be a valuable resource for timely identification of critical ad hoc health-related incidents or serious epidemic outbreaks. In this paper, we explore emotion classification of Twitter microblogs related to localized public health threats, and study whether the public mood can be effectively utilized in early discovery or alarming of such events. We analyse user tweets around recent incidents of Ebola, finding differences in the expression of emotions in tweets posted prior to and after the incidents have emerged. We also analyse differences in the nature of the tweets in the immediately affected area as compared to areas remote to the events. The results of this analysis suggest that emotions in social media microblogging data (from Twitter in particular) may be utilized effectively as a source of evidence for disease outbreak detection and monitoring.

We present the task of predicting individual well-being, as measured by a life satisfaction scale, through the language people use on social media. Well-being, which encompasses much more than emotion and mood, is linked with good mental and physical health. The ability to quickly and accurately assess it can supplement multi-million dollar national surveys as well as promote whole body health. Through crowd-sourced ratings of tweets and Facebook status updates, we create message-level predictive models for multiple components of well-being. However, well-being is ultimately attributed to people, so we perform an additional evaluation at the user-level, finding that a multi-level cascaded model, using both message-level predictions and userlevel features, performs best and outperforms popular lexicon-based happiness models. Finally, we suggest that analyses of language go beyond prediction by identifying the language that characterizes well-being.

Although dietary supplements are widely used and generally are considered safe, some supplements have been identified as causative agents for adverse reactions, some of which may even be fatal. The Food and Drug Administration (FDA) is responsible for monitoring supplements and ensuring that supplements are safe. However, current surveillance protocols are not always effective. Leveraging user-generated textual data, in the form of Amazon.com reviews for nutritional supplements, we use natural language processing techniques to develop a system for the monitoring of dietary supplements. We use topic modeling techniques, specifically a variation of Latent Dirichlet Allocation (LDA), and background knowledge in the form of an adverse reaction dictionary to score products based on their potential danger to the public. Our approach generates topics that semantically capture adverse reactions from a document set consisting of reviews posted by users of specific products, and based on these topics, we propose a scoring mechanism to categorize products as “high potential danger”, “average potential danger” and “low potential danger.” We evaluate our system by comparing the system categorization with human annotators, and we find that the our system agrees with the annotators 69.4% of the time. With these results, we demonstrate that our methods show promise and that our system represents a proof of concept as a viable low-cost, active approach for dietary supplement monitoring.

To support people trying to lose weight and stay healthy, more and more fitness apps have sprung up including the ability to track both calories intake and expenditure. Users of such apps are part of a wider “quantified self” movement and many opt-in to publicly share their logged data. In this paper, we use public food diaries of more than 4,000 long-term active MyFitnessPal users to study the characteristics of a (un-)successful diet. Concretely, we train a machine learning model to predict repeatedly being over or under self-set daily calories goals and then look at which features contribute to the model's prediction. Our findings include both expected results, such as the token “mcdonalds” or the category “dessert” being indicative for being over the calories goal, but also less obvious ones such as the difference between pork and poultry concerning dieting success, or the use of the “quick added calories” functionality being indicative of over-shooting calorie-wise. This study also hints at the feasibility of using such data for more in-depth data mining, e.g., looking at the interaction between consumed foods such as mixing protein- and carbohydrate-rich foods. To the best of our knowledge, this is the first systematic study of public food diaries.

The following sections are included:

• Workshop Focus
• Workshop Agenda
• Workshop Contributions
• References

Technological advances are making large-scale measurements of microbial communities commonplace. These newly acquired datasets are allowing researchers to ask and answer questions about the composition of microbial communities, the roles of members in these communities, and how genes and molecular pathways are regulated in individual community members and communities as a whole to effectively respond to diverse and changing environments. In addition to providing a more comprehensive survey of the microbial world, this new information allows for the development of computational approaches to model the processes underlying microbial systems. We anticipate that the field of computational microbiology will continue to grow rapidly in the coming years. In this manuscript we highlight both areas of particular interest in microbiology as well as computational approaches that begin to address these challenges.

Rare genetic disorders affect millions of individuals worldwide. Many of these disorders can take decades to correctly diagnose. Because of this, genome sequencing of newborns raises a substantial opportunity to identify genetic disorders before they present symptoms, and to identify patient risks at the start of life. Many of these disorders can take decades to correctly diagnose. Because of this, genome sequencing of newborns raises a substantial opportunity to identify genetic disorders before they present symptoms, and to identify patient risks at the start of life. This workshop will report on efforts to screen newborns using genetic sequencing technologies, and attendant biomedical informatics and computational biology approaches.

The use of large-scale data analytics, aka Big Data, is becoming prevalent in most information technology discussions, especially for the life and health sciences. Frameworks such as MapReduce/Hadoop are offered as “Swiss-army knives” for extracting insights out of the terabyte-sized data. Beyond the sheer volume of the data, the complexity of the data structure associated with such data sets is another issue, and may not be so readily mined using only these technological solutions. Rather, the issues around data structure and data complexity suggest new representations and approaches may be required. The LinkedData standard (W3C, Semantic Web) has been promoted by some communities to address complex and aggregatable data, though it focuses primarily on querying the data and performing logical inferences on it, and its use in deep mining application is still in the early stages. In summary, there appears to be a gap between how we access structured data, and the deeper analyses we want to perform on it that preserve representation…

Social media has evolved into a crucial resource for obtaining large volumes of real-time information. The promise of social media has been realized by the public health domain, and recent research has addressed some important challenges in that domain by utilizing social media data. Tasks such as monitoring flu trends, viral disease outbreaks, medication abuse, and adverse drug reactions are some examples of studies where data from social media have been exploited. The focus of this workshop is to explore solutions to three important natural language processing challenges for domain-specific social media text: (i) text classification, (ii) information extraction, and (iii) concept normalization. To explore different approaches to solving these problems on social media data, we designed a shared task which was open to participants globally. We designed three tasks using our in-house annotated Twitter data on adverse drug reactions. Task 1 involved automatic classification of adverse drug reaction assertive user posts; Task 2 focused on extracting specific adverse drug reaction mentions from user posts; and Task 3, which was slightly ill-defined due to the complex nature of the problem, involved normalizing user mentions of adverse drug reactions to standardized concept IDs. A total of 11 teams participated, and a total of 24 (18 for Task 1, and 6 for Task 2) system runs were submitted. Following the evaluation of the systems, and an assessment of their innovation/novelty, we accepted 7 descriptive manuscripts for publication —5 for Task 1 and 2 for Task 2. We provide descriptions of the tasks, data, and participating systems in this paper.