Posters

Search:

View Posters By Category

Session A: (July 7 and July 8)	Session B: (July 9 and July 10)
3Dgeomics (Special Session) CAMDA CompMS Education Evolution and Comparative Genomics Function HiTSeq MLCSB NetBio RNA SysMod TransMed VarI General Computational Biology	3DSIG Bio-Ontologies BioVis CAMDA Evolution and Comparative Genomics HiTSeq MICROBIOME MLCSB RegSys SCANGEN: Single-cell cancer genomics (Special Session) General Computational Biology

Short Abstract: Network clustering aims to identify groups (clusters, communities) of "topologically related" nodes that indicate important functional units within the network. Unlike traditional *static* network clustering, we focus on community detection in *dynamic* networks. In this context, we proposed two novel computational directions, which we then used to study evolution of cellular functioning with age. First, traditional assumption is either that each data time point (e.g., age) has a distinct community organization or that all time points share a single community organization. Instead, we developed SCOUT to divide the dynamic network into contiguous time segments such that all time points in the same segment share a single community organization while time points in different segments have distinct community organizations. In evaluation on dynamic networks from several domains, SCOUT *always* outperformed existing methods. When applied to dynamic network data capturing interactions between proteins at different ages, SCOUT correctly identified as segments the different stages of the aging process (e.g., childhood vs. adulthood). Second, traditional assumption is that nodes should be clustered if they are densely interconnected. Instead, we developed ClueNet to cluster nodes that are topologically similar. In evaluation on the dynamic aging-related network, ClueNet outperformed existing methods 83% of the time.

Short Abstract: Gene expression changes at the mRNA level are induced by rewiring of upstream cell signaling pathways. While gene expression data at the mRNA level are abundant and can be globally and accurately measured, profiling the activity of cell signaling pathways at the proteome level is currently much more difficult. The Expression2Kinases (X2K) algorithm computationally predicts involvement of upstream cell signaling pathways, given a signature of differentially expressed genes at the mRNA level. X2K Web is a web-based implementation of the original Expression2Kinases algorithm with important enhancements. X2K Web includes new transcription factor and kinase gene-set libraries and protein-protein interaction networks. As a case study, thousands of gene expression signatures induced by small molecule kinase inhibitors, applied to breast cancer cell lines, are provided for fetching into X2K Web. The results are displayed as interactive downloadable vector graphic network images and bar graphs. Benchmarking various settings via genetic algorithms (GA) enables the search for optimal parameters to be used in X2K Web to provide the most likely correct upstream regulatory cell signaling subnetwork given sets of differentially expressed genes. X2K is freely available at http://amp.pharm.mssm.edu/X2K

Short Abstract: A SNP is the most well-known genetic variation and causes differences in proteins that make changes in drug efficacy. Here, we use diverse public genetic information and verify the impacts of SNPs in drug efficacy. We make two types of drug action sub-network: without and with SNPs. First of all, we collect PPI network and construct sub-network without SNPs by finding shortest paths from target to pathway genes. Second, to confirm the effects of genetic variants, we gain tag SNPs for each pathway gene and target genes that have regulatory relationships with transcription factors(TF). Based on above PPI network, we construct new PPI network with SNPs for each pathway gene. If TF is lost by tag SNPs, we remove all linkages that are associated with pathway gene. However, if it’s gained by tag SNPs, we insert new connections between pathway gene and target genes that have regulatory relationships. Drug action sub-network with SNPs is built by finding shortest paths based on PPI network with SNPs. Finally, we calculate sub-networks with equations as defined in this study and conclude that positive and negative value represent increase and decrease of efficacy, respectively. We identify that seven of nine cases match with SNPedia.

Short Abstract: Gene regulatory networks (GRNs) describe regulatory relationships between transcription factors and their target genes. Computational methods to infer GRNs typically combine evidence across different experiments and conditions to infer context agnostic networks. However, many applications, including Drosophila sex differentiation, require contexts specific networks. We present a novel method Network Reprogramming using EXpression (NetREX) that constructs a context specific GRN given context specific expression data and a context agnostic, or incomplete prior network. NetREX approach is unique in its exploration of the landscape of possible GRN topologies converginge on the best explanation for expression data, subject to several biologically motivated modeling constraints. Because NetREX rewards topologies related to prior network, we also developed and validated, PriorBoost a method that evaluates a prior network in terms of its consistency with the expression data to determine if reconstruction is likely to be successful. We applied NetREX to construct the first sex-specific Drosophila GRNs. We validated the predicted GRNs computationally and experimentally and compared the results obtained from NetREX with those obtained from alternative methods. NetREX constructed sex-specific Drosophila GRNs that, on all applied measures, strikingly outperformed networks obtained from other methods indicating that NetREX is an important milestone towards building more accurate GRNs.

Short Abstract: Synthetic lethality (SL) occurs between gene pairs and is the phenomenon where the inactivation of individual genes has no effect on cell survival, but their co-inactivation results in cell death. Though several studies exploit SL interactions to develop patient specific therapeutic options, in vitro and in vivo experiments are time and resource intensive and focus primarily on known cancer associated genes. Therefore, computational methods are being developed to identify potential SL interactions worth exploring in vitro. Here we develop a cancer independent humanized yeast genetic interaction network (HYGIN) based on experimentally validated yeast orthologs to identify novel patient specific gene targets for personalized medicine. Using one-to-one ortholog mapping (yeast to human) our unbiased approach generated a humanized yeast genetic interaction network that contains 1,009 genes and 10,419 interactions. The network was validated with previous in vitro and in silico approaches. Breast cancer gene expression data was then applied to the network to generate a sub-network to identify SL interactions that are breast cancer specific. A total of 15 genes that are significantly down-regulated in breast cancer at a 2-fold cut off have a total of 115 SL partners that can be further explored for drug interaction.

Short Abstract: Whole exome sequencing (WES) studies for Autism Spectrum Disorder (ASD) could identify only around six dozen risk genes to date because the genetic architecture of the disorder is highly complex. To speed the gene discovery process up, a few network-based ASD gene discovery algorithms were proposed. Although these methods use static gene interaction networks, functional clustering of genes is bound to evolve during neurodevelopment and disruptions are likely to have a cascading effect on the future associations. Thus, approaches that disregard the dynamic nature of neurodevelopment are limited. We present a spatio-temporal gene discovery algorithm for ASD, which leverages information from evolving gene coexpression networks of neurodevelopment. The algorithm solves an adapted prize-collecting Steiner forest based problem on coexpression networks to model neurodevelopment and transfer information from precursor neurodevelopmental windows. The decisions made by the algorithm can be traced back, adding interpretability to the results. We apply the algorithm on WES data of 3,871 samples and identify risk clusters using coexpression networks of early- and mid-fetal periods. On an independent dataset, we show that incorporation of the temporal dimension increases the predictive power: Predicted clusters are hit more and show higher enrichment in ASD-related functions compared to the state-of-the-art.

Short Abstract: A network motif is defined as an overly represented unique subgraph pattern in a network and has been applied in various biological and medical problems. Network motif detection algorithms are categorized into network-based and motif-based approaches and various tools are currently available. However, most existing tools are outdated, incompatible with modern operating systems, or fail to provide sufficient operation instructions. Additionally, most tools provide limited information regarding network motifs, which necessitates post-processing program to apply to real problems. Consequently, the lack of usability brings a certain amount of skepticism about the relevance of network motifs in investigating real biological problems. Recently, NemoProfile is introduced as an efficient new network motif data model which eliminates the need for costly two-step processing’s. In this paper, we introduce NemoLib (network motif library) as a general purpose tool for detection and analysis of network motifs, which easily implement various network motif analyses and applications. Currently, NemoLib is available in C++, Java and python languages, providing for extensibility. We demonstrate how NemoLib is used for visualizing networks and network motifs, combining network-based and motif-based approaches, producing NemoProfile, and parallelizing with MASS library.

Short Abstract: Signaling pathways are a core focus in systems biology to understand how individual proteins and their interactions contribute to a larger cellular response. Tools to automatically reconstruct signaling pathways from protein-protein interaction (PPI) databases can help biologists generate testable hypotheses about signaling. PathLinker is a reconstruction tool that connects pathway-specific receptors to known downstream pathway-specific transcriptional regulators within a weighted interactome using a shortest paths approach. While PathLinker showed promising results, we observed that a substantial number of the returned paths share the same reconstruction cost due to ties in the interactome edge weights. Further, the paths are not all biologically relevant to the pathway of interest. Through a dynamic programming approach, we extend PathLinker by incorporating information about the cellular localization of the interacting proteins to ensure that the individual proteins are localized in cellular compartments involved in signaling transduction, and that the PPIs are spatially coherent with the signaling flow, which typically starts at a membrane-bound protein receptor and is transmitted downstream through a succession of PPIs within the cytosol to end at a transcription regulator inside the nucleus. Results show that the proposed method leads to more accurate and biologically meaningful reconstructions on benchmark datasets of PPIs.

Short Abstract: Acute Promyelocytic Leukemia (APL) is caused by a differentiation block in the hematopoietic myeloid lineage cascade, which leads to accumulation of immature granulocytes (promyelocytes) in the blood. Traditional APL therapy includes all-trans retinoic acid (ATRA). This approach is short-term because the cells quickly become ATRA-resistant. Combination of ATRA with chemotherapy is required to yield transient cancer remission. Here, we leverage on the reprogramming power of the transcription factors (TFs) that control the granulocyte-to-neutrophil differentiation as an alternative to chemotherapy. The effect of ATRA and DMSO treatment on NB4 APL cells was monitored in a time-series RNASeq experiment. The differentially expressed (DE) genes between ATRA and DMSO-treated cells of each time point were calculated, considering the effect of time on gene expression. The most drastic differences happen after 4h and 72h of treatment. We applied a computational network approach (mogrify) to identify the TFs which control the DE genes at these time-points. Correlation of these TFs with their targets suggested the most promising combinations of TFs that simulate the ATRA induced phenotype. We identified some known TFs (GATA2, CEBPB, MYC) as well as some novel ones (IRF1, TGIF1 and others) which represent potential targets for therapeutic modulation in leukemia cells.

Short Abstract: Circulating tumor cells (CTCs) are an emerging type of liquid biopsy that enables minimally invasive monitoring of progression and metastasis of prostate cancer. Though advances in single-cell RNA-Seq (scRNA-Seq) technology potentially provide a cost-effective way to understand the transcriptomic landscape of CTCs, data sparsity has greatly limited its application to the inference of differential gene regulatory networks modulated by cellular states of individual CTCs. Here we devised a differential network analysis that normalizes gene-gene correlations based on the sparsity of scRNA-Seq reads data and statistically infers the changes in correlations between groups of CTCs. Applying the method to analyze a dataset of 169 prostate CTCs, we built a highly intertwined gene regulatory network rewired during different cell cycle phases. Functional annotation analysis of the network revealed significant associations to crucial functions governing cancer progression and metastasis. For instance, IL17RA, a well-studied player of prostate tumor growth and invasion, was a hub genes in the network that regulated a significant number of genes specifically in the synthesis phase. In summary, our method realized the analysis of differential gene regulatory networks using highly sparse scRNA-Seq data and illuminated crucial biological functions modulated by the cell cycle of prostate CTCs.

Short Abstract: With the rapid depletion and deleterious effects that non-renewable energy sources (such as petrol, coal, and diesel) are having on the environment, there is a momentum to utilize cleaner, renewable energy sources. Bioethanol is one such renewable energy source that is applicable to the needs of Indian society. In the last few years, research into the use of 2nd generation lignocellulosic biomass as a feedstock for the production of bioethanol has gained considerable momentum. [1]. However, enzymes that degrade the recalcitrant crystalline cellulose at an industrial level has always been a barrier in efficient production. Other than Neurospora crassa and Clostridium thermocellum, other fungal and bacterial organisms that are viewed as potential sources of cellulolytic enzymes are less studied in terms of their metabolic pathways and the interplay of cellulolytic enzymes involved in cellulose degradation [1]. Therefore, to address this problem, we have performed a network driven computational-based systems meta-analysis on all the ChiP-seq data available for fungal species known to have lignocellulosic decomposition characteristics. This analysis would help us to identify novel genes being co-expressed along with known genes (like lytic polysaccharide monooxygenase [2]) thereby providing new insights into ways to improve hyper-production strains.

Short Abstract: Identification of modules in molecular networks is at the core of many current analysis methods in biomedical research. However, how well different approaches identify disease-relevant modules in different types of networks remains poorly understood. We launched the “Disease Module Identification DREAM Challenge”, an open competition to comprehensively assess module identification methods across diverse gene, protein and signaling networks. Predicted network modules were tested for association with complex traits and diseases using a unique collection of 180 genome-wide association studies (GWAS). While a number of approaches were successful in terms of discovering complementary trait-associated modules, consensus predictions derived from the challenge submissions performed best. We find that most of these modules correspond to core disease-relevant pathways, which often comprise therapeutic targets and correctly prioritize candidate disease genes. This community challenge establishes benchmarks, tools and guidelines for molecular network analysis to study human disease biology (https://synapse.org/modulechallenge).

Short Abstract: C. elegans is a perfect model organism for aging studies due to its short lifespan and expansive genetic information available. Studies on the model have uncovered key ageing pathways, along with revealing to us the staggering complexity of the process. The complexity, thus, invites a top-down systems approach to pinpoint central systems affecting ageing. Our goal is to characterize the complex systems of gene expression during early adulthood in C. elegans utilizing network inference. With timely-resolved (at four hour intervals) transcriptomics data of long-lived and normal-lived C.elegans, along with the wealth of –omics data publicly available, we integrate the data to construct an evidence-based gene regulatory network. Next is to identify promising algorithms that infer causal relationships to best depict the crucial moment triggering ageing process. Recent developments have shown that utilizing wisdom of the crowd and incorporating extensive priors information vastly increases the accuracy of the predictions. Our efforts aim at resolving the first changes that begin the ageing process in worms and to later also understand if similar molecular mechanisms influence ageing in humans.

Short Abstract: Non-negative matrix factorization is a popular tool for decomposing high-dimensional data into its constitutive parts. Recent research has incorporated manifold regularization to select parts which are consistent on a manifold, a mathematical structure that describes how features in the data are related to one another. One source of hidden structure in genomic data relates to the effect of interactions among groups of genes in the cell. Applications of manifold-regularized matrix factorization to these datasets have revealed cancer sub-types with different biomarkers and effective treatments. We propose a more focused version of this method which reformulates the global gene interaction network manifold as a set of sub-manifolds, each associated with a biological pathway. Biological pathways are important sets of interactions because they define how pairs of interacting genes influence broader physiological processes. These processes also capture tissue- and context-specific relationships among the genes under investigation. By constraining matrix factorization techniques to respect these underlying structures and emphasizing pathway edges instead of nodes, we gain biological insight when examining the factors.

Short Abstract: Identification of essential genes specific to a particular cancer cell-line is crucial to find a proper therapeutic strategy to treat cancer. Topological analyses of protein-protein interaction (PPI) network have been applied for the discovery of essential genes. For example, hubs in a network tend to have a higher chance to become an essential genes which establish the centrality-lethality (C-L) rule. However, current analyses often do not consider tissue-specific changes of gene expression so that topology analyses specific to cancer cell-types were not applicable to predict cancer essential genes. Here, we construct the “cancer cell-line specific PPI network” through integrating PPIs with gene pairs co-expressed on cancer cell-line transcriptomes. We discover that newly identified central nodes in the cancer cell-line specific networks are different from those in the preexisting network and identify more cancer essential genes. Our study provides shed light on targeting genes for precise cancer therapeutics based on transcriptomic profiles of cancer cell types.

Short Abstract: Orthologous genes generally perform conserved functions across species, and are thus associated with same or closely similar phenotypes. Based on the notion, mice have been widely used as a model organism to investigate human gene-phenotype relationships. However, phenotypes associated with orthologous genes often turn out to be quite different between human and mouse. To overcome this difficulty, we investigate biological descriptors to find orthologous genes of similar phenotypes between human and mouse. With the help of available phenotypic data, we separated orthologous genes into two groups, 642 genes of similar phenotypes between human and mouse models and 642 genes of different phenotypes. We compared expression dynamics in network partners of the two groups, and found that phenotype similarity and differences can be better explained by a group of genes in the interaction network than a single gene analysis, which implies cooperativity of genes may determine phenotypic consequences. Orthologs of different phenotypes between human and mouse tend to show diverse gene expression changes of network partners compared to orthologs of similar phenotypes. Our study improves understanding of phenotypic changes in orthologous genes and provides an opportunity to select genes suitable for model organism research applicable for certain human disease phenotype.

Short Abstract: Druggable protein-protein interactions (PPIs) has been considered as efficient drug targets, which could be disrupted by small molecules or ligands, and affects specific pathways and causes less side effects than targeting whole protein as drug target1,2. To accurately predict druggable PPIs, we need to have protein 3D structures which provide the information for PPI interface and potential drug binding region3. However, among 41,000 human PPIs, only ~2,500 nonredundant protein complex structures available in Protein Data Bank4,5. Therefore, a method to find druggable PPIs is required to resolve current limitations from the limited protein structural information. We utilized domain-mediated interaction (DMI) as a novel information to find druggable PPIs, since DMI interfaces are known to associate with chemical binding pockets and help proteins perform specific function. We identify relationships between DMI interfaces and chemical binding pocket, and predict druggable PPIs using them. Then, we show various literature evidences of the predicted druggable PPIs. Next, we classify druggable PPIs into monotropic- and pleiotropic-druggable PPIs. We further confirm positive correlation between the numbers of associated DMI interfaces and of related side effects of drug target PPIs. Our results suggest that DMI interface provides insight for identifying druggable PPIs with minimal structural information.

Short Abstract: Integration of biological networks into data analyses are common techniques within bioinformatics workflows, but still face problems in reproducibility, particularly in situations that build upon collaborative work. Within the process, the visualization of networks has huge impact on the interpretation, and therefore is an essential step in communicating even intermediate results. Furthermore, the shift of bioinformatics towards systems medicine, and its application within a clinical setting demand the establishment of interconnected and interchangeable components and, moreover, a standardized interface for information exchange with healthcare systems. We demonstrate a course from data acquisition to the visualization, while providing an interface for healthcare systems by exposing Fast Healthcare Interoperability Resources (FHIR®). Thereby the network data exchange (NDEx) platform and the Cytoscape project form the core components. We use our R package ndexr to retrieve networks from public and private NDEx installations, and also to store the results for collaboration and publication. Cytoscape is a major tool for the visual exploration of biomedical networks. Beside the graphical interface, it can be accessed by the R package RCy3. To reduce the complexity of software installations, we utilize docker to compose the different components, illustrating their interchangeability and flexibility furthermore.

Short Abstract: Background Parkinson’s disease (PD) is the second most common neurodegenerative disorder after Alzheimer’s disease. Despite extensive but separate analyses of gene and miRNA expression in PD, the regulatory interplay between deregulated genes and miRNAs triggering PD remains largely uncharacterized. We present here a comprehensive integrative analysis of gene expression profiles, miRNA signatures, and other publicly available regulatory databases in order to better understand the collaborative functional role between miRNAs and genes in driving PD. Results A PD-specific transcription factor (TF)-miRNA regulatory network was generated and its topology and functional impacts upon PD was analyzed. We identified six driver genetic elements (2 genes and 4 miRNAs) and various functional network modules that could conceivably trigger PD etiology. Moreover, neuroprotective agents and other small molecule signatures were predicted to reverse the transcriptional changes caused by the identified functional modules. Conclusion Our joint regulatory analysis of coding and non-coding RNA, has the potential to yield clinically as well as biologically relevant information. These results, on the one hand, enrich our knowledge base of prospective genomic drivers and essential network modules in PD progression and, on the other hand, open new opportunities towards novel therapeutic leads and better treatment for PD patients.

Short Abstract: Human immunology relies heavily on high-throughput data to infer the underlying mechanisms. A common set of computational tools are needed to model the immune responses to infectious diseases, vaccines, autoimmune diseases, inflammation and cancer immunotherapy. Building upon our previous works on blood transcription modules for transcriptomics analysis, and mummichog for metabolomics pathway/network analysis, we have developed a method of multi-omics, multiscale networks to integrate big data immunology. The approach combines meaningful dimension reduction, hierarchical networks and partial least square regression to explain individual differences in immune response. We will demonstrate the application to a herpes zoster vaccine study (Li et al, 2017. Cell 169:862), where important metabolic phenotypes are identified through the integration of transcriptomics, metabolomics, cytokines and cell populations. New results will be reported on a study of controlled malaria infection in humans, and on continued effort in software development.

Short Abstract: Despite their ubiquity and obvious importance, our level of understanding of viruses generally is still equivalent to being in the dark ages. In fact, most viruses sequenced have no resemblance to any characterized known virus. Recently, we developed a new approach for characterizing novel viral DNA sequences using graph and network theory. In this network, each node corresponds to a gene family and nodes are connected if they are found within the same viral genome sequence. As current sequencing costs continue to decrease while throughput increases, genomic data is being generated at a rapid pace. While this new data can be integrated into our existing viral gene networks by recomputing the entire network, this solution is computationally intensive. Thus, we have developed a heuristic method for inserting new data sets into our existing gene network. This method determines the placement of new sequence data based upon the Jaccard index for k-mer sequence compositions. We have benchmarked this approach; in our current implementation we can accurately place new sequence data within our network with at least 80% accuracy.

Short Abstract: Differential network analysis that highlights interaction changes between conditions is a rising paradigm in network biology. However, differential network analysis is usually applied to compare between few conditions, and a large-scale assessment of different methods has been lacking. Here, we developed approaches that extend this paradigm toward differential network analysis of multiple conditions, which we applied to analyze protein interaction networks (interactomes) of 34 human tissues. We assessed the performance of these approaches by their ability to expose physiological differences among human tissues relevant in health and disease. To this end, we manually associated 6,019 biological processes and 1,185 hereditary diseases with their relevant tissue. Our analysis revealed the advantage of differential tissue interactomes over node-weighted tissue interactomes in highlighting tissue-specific processes. We then tested whether differential tissue interactomes can highlight tissue-selective disease-related genes. Indeed, in 28% of the 1,527 cases we studied, the top 1% differential sub-network surrounding a causal gene was enriched significantly for additional disease-causing genes, a fraction much higher than expected by chance. Altogether, the expansive resources that we created and our analyses demonstrate that differential analysis of human tissue interactomes is a powerful tool for highlighting processes and genes with tissue-specific functionality and impact.

Short Abstract: Biological network alignment (NA) aims to uncover similar regions between protein-protein interaction (PPI) networks of different species. Then, analogous to genomic sequence alignment, NA can be used to transfer functional knowledge across species between their conserved PPI network (rather than sequence) regions. For example, if we align the PPI network of baker's yeast, a well-studied species, to the PPI network of human, a poorly-studied species, we can infer the function of human proteins based the function of their aligned partners in the yeast network. Existing NA methods are homogeneous, i.e., they can deal only with networks containing nodes and edges of one type. However, analyzing heterogeneous multi-node- or multi-edge-type network data (such as genes/proteins, phenotypes, diseases, or drugs, and various types of interactions that can exist between them) can lead to deeper insights into cellular function. So, we introduce several algorithmic novelties to allow three recent state-of-the-art NA methods, WAVE, MAGNA++, and SANA, to perform heterogeneous NA for the first time. In evaluations on synthetic, PPI, and protein-Gene Ontology term networks, our proposed heterogeneous NA methods lead to higher-quality alignments and better robustness to noise in the data than their homogeneous counterparts.

Short Abstract: The precision medicine paradigm is centered on targeted therapies, or drugs designed to interact with particular molecular entities and elicit an anticipated and controlled therapeutic response. To better understand the effects of targeted therapies in patients, we need computational approaches, which are inadequate at present, to map drug targets into a picture providing greater context, such as pathways or networks. To address this need, we have developed a user-friendly intuitive computational tool by enhancing a widely used Cytoscape app, ReactomeFIViz, to assist researchers to visualize, model, and analyze drug and target interactions by facilitating high quality manually curated pathways and a genome-wide human functional interaction network from Reactome, the most comprehensive open source biological pathway knowledgebase. We describe several examples demonstrating the application of our tool for visualizing cancer drugs in the contexts of pathways and networks, and simulating the effect of drug perturbation on pathways based on fuzzy logic models. This pathway- and network-based tool will allow researchers to identify subgroups of patients with shared dysregulated pathways, the commonalities that can provide greater understanding into patient subgroup attributes such as drug sensitivity or resistance, drug toxicity or adverse side effects, and potential opportunities for targeting by combination therapies.

Short Abstract: Reactome (https://reactome.org) is one of the few open access curated biological pathway knowledgebases. Its authoritative and detailed content has directly and indirectly supported basic and translational research studies with over-representation analysis and network-building tools to discover patterns in high-throughput data. With user experience firmly in mind, we have created a new web interface. An improved version of the Pathway Diagram Viewer provides faster data loading, diagram rendering and element seeking. Furthermore, to increase the quality of the graphics used to represent pathways, we have integrated Enhanced High-Level Diagrams (EHLDs). These illustrations look like overviews of biological processes found in textbooks and review articles, with a consistent iconography with embedded navigation and data overlay functionality. We have expanded our current Probabilistic Graphical Models (PGM) for Reactome pathways to support inference of pathway impacts of somatic mutations affecting protein structures in tumours. We’ve developed new factor functions to integrate drug impact on pathway activities. We have established an approach to convert Reactome pathways into Boolean networks (BN) automatically and perform pathway simulations to infer drug and somatic mutation impact. Both PGM- and BN-based pathway modelling approaches and new user-intuitive visualization features have been implemented in ReactomeFIViz, the Reactome Cytoscape app.

Short Abstract: Autism spectrum disorder (ASD) has a high comorbidity with other developmental and neuronal disorders, including congenital heart disorder (CHD) and epilepsy (EPI). The molecular and genetic basis of these disorders and their shared underlying mechanisms causing the observed co-morbidity are not fully understood. We postulate that ASD, CHD, and EPI are diseases of networks, not genes, and the abundance of shared phenotypes between pairs of disorders suggests that deeper understanding of the molecular mechanisms involved may be gained by examining the networks at the intersection of each pair. With the help of network propagation and tissue-specific interactomes, we prioritize de novo variants (DNVs) in three cohorts. We find an overabundance of DNVs in network proximity to the high confidence genes, confirming that many are related to the disorder, as well as identifying novel disease-gene candidates. In addition, a novel technique of paired network propagation identifies DNVs at the intersection of ASD and CHD high confidence genes. Further cross-referencing the ASD-CHD prioritized genes with an independent dataset confirms that these genes successfully associate with the combined phenotype. These findings demonstrate that a network-based framework is crucial for uncovering the shared molecular mechanisms of neurodevelopmental disorder pathogenesis.

Short Abstract: Recent improvements in the throughput of chemical-genetic interaction screens have necessitated the development of scalable computational methods for prioritizing candidate targets derived from chemical-genetic screen data. We developed a computational pipeline to integrate multi-modal interactions measured as hypersensitivity of heterozygous mutant diploids as well as haploid gene deletion or temperature-sensitive mutants to bioactive compounds and prioritize the most significant compound-gene interactions in multiple modes. We incorporated a two-sided analysis, compound-wise and gene-wise, where the most specific outliers against a compound or a gene, respectively, were evaluated. We applied this pipeline to the chemical-genetic interaction data of about one thousand compounds from the RIKEN Natural Product Depository and other compound libraries against S. cerevisiae mutant collections and identified a number of compounds with highly specific putative targets. We confirmed the accuracy of these novel targets by generating drug resistant mutants and utilizing whole-genome sequencing to identify genetic loci where resistance-associated mutations emerged. Overall, this pipeline provides a strong foundation for identifying targets of bioactive compounds from several diverse chemical-genetic experiments and facilitates the discovery of novel targets for previously uncharacterized compounds.

Short Abstract: Since the early 2000s, network modeling analysis has become a popular tool for identifying functional modules from networks built using high-throughput biological data. Networks are relatively easy to build, aesthetically pleasing, and typically offer a novel approach to modeling large heterogeneous relationship sets, making them a very useful tool. However, networks are vulnerable to irreproducibility. There are multiple approaches to network building, compounded with a lack of well-defined measures for quality control. Considering the increasing use of networks in modeling various types of biological relationships, examining the reliability and robustness of biological networks represents one of the most critical problems in biomedical research. In this research, we investigate the reliability of the correlation network model by perturbing gene correlation networks from Mus musculus to determine effect on structure and resulting biological function. We find that high edge-density regions, or clusters, are more robust to perturbation than low-density regions. This research indicates that correlation models preliminarily appear to have varied reliability, where high density regions tend to be more structurally and biologically stable, and low density regions are more prone to being critically affected by noise. These results demonstrate that more studies into the robustness of the network model are needed.

Short Abstract: Tumor heterogeneity is the leading cause of drug resistance and metastatic progression. One aspect of tumor heterogeneity is driven by interactions between different cell types within the tumor microenvironment that enable malignant cells to adapt and proliferate. An active research area is to identify cell-cell signaling interactions that are specific to tumor maintenance. To date, few such cell-cell signaling interactions have been found for therapeutic interactions. Analyzing the tumor microenvironment by dissociating the tumor-stromal microenvironment and generating transcriptomic data by cell type promises to reveal these cell-cell interactions. Current network-based analytic approaches that leverage such data to identify TME-specific interactions are largely based on correlation analysis, but these can lead to high false positive rates. Instead, we aim to identify potential ligand-receptor interactions by enriching for consistent downstream biological processes. We developed a novel network-based algorithm to impute these interactions that captures the effect of various cell types interacting simultaneously. This algorithm was applied to RNA-Seq data from individually sorted stromal and epithelial cells from the primary human non-small cell lung cancer. The results show both novel, and confirm known, TME interactions, aid with the interpretation of prognostic signatures, and enable identification of potential tumor-stromal interaction drug targets.

Short Abstract: An important aspect of studying human diseases and developing new treatment strategies is the use of animal models. However, it is crucial to take into account the intrinsic differences in regulation between human and animal models as well as the mechanisms underlying the specific phenotype in order to generate reliable hypotheses. Moreover, we still need to overcome the challenge of identifying the regulatory genes and pathways that are responsible for a phenotype of interest both in human and in animal models. Therefore, we have designed a unique comparison framework for investigating pathways in mammalian organisms on different levels of detail based on data integration and network analysis techniques. The orthology relationships between different organisms as indicated by the eggNOG database (http://eggnogdb.embl.de) were used to identify and visualize the pathway conservation. We show that all KEGG pathways overlap more than 50% between human, mouse, rat and pig. Furthermore, we integrated these pathways with tissue expression data from the recently updated TISSUES database (http://tissues.jensenlab.org/) to enable the identification of expression patterns across mammalian tissues and organisms. Finally, we discuss how our comparison framework can be used to suggest the most suitable model organism for the relevant human disease (sub)pathways.

Short Abstract: Despite the prevalence of genome-wide association studies, it remains a challenge to identify the causal variants and genes that drive complex diseases. Susceptibility to such diseases may arise via distinct mechanisms in different individuals, and involve multiple low-penetrance factors working together. To address this problem, network-based methods prioritize candidate genes by evaluating their role in the network of molecular interactions in the cell. However, these methods often rely on static interaction networks and do not take into account how network structures change across different conditions, diseases, or tissues types. We recently developed an algorithm called ALPACA (ALtered Partitions Across Community Architectures) to find differences in modular structure between condition-specific networks. Here, we apply ALPACA to compare regulatory networks in patients with chronic obstructive pulmonary disease (COPD) and matched healthy controls. We find that known COPD risk genes tend to cluster together in regions of the network that are structurally altered in COPD patients, rather than in the densest modules across the whole network. Our results suggest that functional variants reorganize regulatory networks to alter their modular structure, and that examining disease-specific changes in network structure could help us identify potential causal genes and ultimately discover novel therapeutic strategies.

Short Abstract: The Human epidermal growth factor receptor 2 (ERBB2/ HER2-Neu) has expanded expression in 15% of breast cancer samples. HER2-enriched samples are associated with poor prognoses. Resistance to therapy with the HER2-targeting monoclonal antibody, Trastuzumab, remains a challenge in treatment of HER2-enriched breast cancers. We explore the transcriptional regulation of HER2 expression across intrinsic subtypes of breast cancer to help identify targets for adjunctive therapy. The PAM50 algorithm was applied to data from breast invasive carcinoma (BRCA) RNAseq version 2 samples from the Cancer Genome Atlas (TCGA). Using the Algorithm for the Reconstruction of Accurate Cellular networks (ARACNe), transcriptional regulatory networks were constructed using basal-like samples only, HER2-enriched samples only, luminal A samples only, luminal B samples only, and all samples. GEne Network Inference with Ensemble of trees (GENIE3), as well as Passing Attributes between Networks for Data Assimilation (PANDA) were also used for that purpose. In HER2-enriched samples, HER2 is uniquely dependent on E2F2, HOXA5 and CRY2. The results bespeak differences in regulation of HER2 across subtypes. Activities/expressions of E2F2, HOXA5, and CRY2 are thus candidates for targeting as potential adjunctive therapy in the treatment of Trastuzumab-resistant HER2-enriched breast cancers.

Short Abstract: Transcriptional regulatory networks (TRNs) are enriched for certain network motifs. This could either be the result of natural selection for particular hypothesized functions of those motifs, or it could be a byproduct of patterns of mutation and more general types of selection. We have developed a powerful new method for distinguishing between the causes, by simulating TRN evolution under different conditions. Our model is the first model of TRN evolution to simulate mutations to transcription factor binding sites with realistic mechanistic details, thus capturing a realistic rate of turnover of links in the TRN. Our simulation of gene expression is also highly mechanistic, capturing stochasticity and delays in gene expression. We use the model to study a well-known motif, the type 1 coherent feed-forward loop (C1-FFL), which is hypothesized to filter out short spurious signals. We found that functional C1-FFLs evolve readily in TRNs under selection for this function, but not in a variety of negative controls. Negative controls do evolve non-functional C1-FFLs, albeit fewer than the functional C1-FFLs promoted by the selection. Interestingly, we found a new “diamond” motif also emerged as a short spurious signal filter in the simulation.

Short Abstract: Metabolism’s chemical reactions interconvert small metabolites and assemble them into larger cellular structures to sustain life’s diverse processes. While omics technologies produce descriptive measurements of metabolism, scientists need context to design experiments and interpret these measurements. Current tools simplify metabolism by dividing metabolites and reactions into distinct subsystems such as glycolysis, pentose-phosphate pathway, or citric acid cycle. These representations distort relations between subsystems, obstructing the study of metabolism as a cooperative, continuous system. We present a strategy and prototypical tool for dynamic exploration of relations within metabolism and for visualization of experimental measurements in this context. Beginning with the latest compilation of knowledge on human metabolism, Recon 2M.2, we curated this information for clarity and enhanced its external references in order to associate experimental measurements to metabolites. Our tool allows users to define custom metabolic networks from this information, with filters for compartments and subsystems of interest, representations with or without compartmental distinction of metabolites, and reduction of excessive connectivity from prolific metabolites or reactions. Our tool also enables users to query these networks by topological traversals and to visualize these networks in dynamic, interactive diagrams. Development is in progress on our application and its interactive interface.

Short Abstract: The molecular basis and dynamics of the plant-pathogen interactions are poorly understood aspects of plant pathogenesis. Here we propose a network tomography approach to identify host-pathogen interactions using multiple types of molecular screens and a network identification methodology inspired from the analysis of computer internetworks. To characterize the events at the plant signaling/pathogen effector interface, we assembled, using a high-throughput protein-protein interaction screen, a host-pathogen network between 139 tomato kinases and four effectors from the bacterial pathogen Pseudomonas syringae. A subset of 36 multi-effector interacting kinases belonging to various structural classes is selected for in-depth phenotypic analyses. We developed a method of pathway inference from phenotypic data using a co-occurrence pattern analysis algorithm. We used the co-occurrence of nodes with immune related phenotypes in multiple network perturbation assays (silencing of effector-interacting kinases) to evolve the host pathogen interaction network subject to architectural contraints. Our analysis identified kinases acting as positive or negative regulators of the innate immunity, effector-triggered immunity (ETI), and programmed cell death (PCD). We identified “essential” kinases for immune responses in the resulting signed networks and integrated the phenotypic data to construct hierarchical stimulus-specific networks. These results provide a framework for prioritizing targets for enhancing plant resistance.

Short Abstract: We present a novel data structure, the cycle-free paths graph (CFPG), for representing and analyzing sets of paths in biological networks. The CFPG is a graphical structure which exactly represents the paths of a given length between two nodes in a directed network without explicit enumeration. Many properties of the ensemble of paths can be characterized through operations on the CFPG. These include counting paths, sampling under different biological assumptions about edge probabilities, or sampling for paths satisfying a Boolean combination of basic constraints of the form: (i) the path passes through the node v (ii) the path does not pass through the node w. We illustrate the use of CFPGs to identify candidate mechanisms for the effects of anti-cancer drugs on a large set of downstream targets in a previously published phospho-proteomic dataset. We generated CFPGs from a diverse set of network representations of causal mechanisms, including a simple directed protein interaction graph, a BEL knowledge graph, and the influence map of an executable Kappa model. Through the construction and analysis of CFPGs we were able to characterize the properties of paths linking drug targets and and effectors in networks where explicit path enumeration would be intractable.

Short Abstract: Screening chemical compounds against genome-wide mutant arrays identifies genetic perturbations that cause sensitivity or resistance to compounds of interest. We developed a chemical-genetic interaction screening pipeline in Saccharomyces cerevisiae to functionally annotate chemical compounds in a high throughput manner. This pipeline leveraged a diagnostic collection of ~300 genetically-barcoded mutants in a drug-sensitized background, a high degree of sample multiplexing, and a genome-wide compendium of genetic interaction profiles to predict modes of action for compounds across seven diverse compound libraries. These libraries contained natural products and derivatives, combinatorially-synthesized compounds, and several hundred approved drugs. Two key components of our pipeline were the computational tools we developed to score chemical-genetic interactions from pooled screens of barcoded mutants and to integrate chemical-genetic and genetic interaction datasets for mode-of-action prediction. We used the latter component to identify a set of ~1500 out of the ~14,000 screened compounds that possessed high-confidence mode-of-action predictions, which we experimentally validated across multiple diverse functions. More broadly, we observed functional biases in the different compound libraries, with the larger and presumably least structurally- and mechanistically-biased collections showing a depletion for compounds perturbing nuclear processes.

Short Abstract: Due to individual’s subtle biological differences and cancer heterogeneity, each cancer patient responds to chemotherapy differently. It is crucial for physicians to identify the drug target, which effectively responds to the right patient at the right cancer stage. We try to answer the question of what key interactions of signal transduction play important roles in the traits of cancer in specific stages. Here, our hypothesis is that differential analysis of pathways in different stages can reveal potential alternative drug targets in stage-specific cases. To do that, we collected RNA expression dataset of 212 lung adenocarcinoma patients and 15 normal cases from TCGA. Second, we performed differential expression analysis based on cancer stages to identify the significantly differential expressed (DE) genes in each stage. Using stage-specific DE genes, we applied RDF Sketch tool that decomposes the integrated pathways into relevant paths of signal transduction, which were reintegrated into stage-specific networks. Then, we applied our published method for differential pathway analysis (REDESIGN) to pinpoint the differential paths resulting in the hallmarks of cancer in each stage. The resulting paths contain potential drug targets in stage-specific cases, so that the experts can take such knowledge into clinical consideration for more effective therapeutic regimen.

Short Abstract: NDEx Common (NC) is a draft network format to promote interoperability of biological networks from diverse sources and communities. NC is a response to the challenges posed by NDEx, the Network Data Exchange (www.ndexbio.org), where biological networks of many types, sizes, and formats can be stored, shared, published. To serve a very broad community, there are no constraints on the schema or semantics networks stored in NDEx. Nevertheless, users want interoperability in order to easily combine networks from different authors and to fuel to their applications with data from many sources. Further, researchers typically want conceptually straightforward models, ones that can be processed in tabular forms or used with standard network tools such as Cytoscape. NC is therefore designed to be a middle ground between sophisticated knowledge representations standards such as BioPAX or BEL and simple formats with ad hoc semantics. It is modular: lower levels specify network annotations and handling of identifier namespaces, higher levels enable detailed representation of biological relationships, organized so that authors can adopt subsets appropriate to their needs. The mission of NDEx Common is to bridge between diverse communities using networks and to facilitate the use of networks in workflows and applications.

Short Abstract: Candidate gene prediction unravels genes with important molecular functions, diseases, and phenotypes, which is a critical step to understand the role of genes and a web of interactions formed by them in a living system. Biological network-based candidate gene prediction is popular but suffers from low prediction accuracy. We developed an integrative framework that combines protein-protein interaction (PPI) networks with anatomy ontology data to increase the candidate gene prediction accuracy. Anatomy ontologies can be used to extract the experimental knowledge about genes and their relationships to anatomical phenotypes. Using semantic similarity, we developed a method to generate protein networks based on the Uberon anatomy ontology and integrated them with the PPI networks downloaded from the STRING database. We predicted the candidate genes for mouse and zebrafish, and we demonstrated that the integrated networks significantly increased the candidate gene prediction accuracy compared to the PPI networks. It is evident that the integration increases the quality of the original PPI networks by mitigating the effects of false positive and false negative interactions. This integrative approach will be useful in predicting candidate genes associated with important phenotypes and diseases.

Short Abstract: PD is the second most common neurodegenerative disease and is characterized by both motor and non-motor symptoms. In a minority of cases, around 5-10%, PD is caused by inherited mutations in single genes. The rest of PD cases, however, are not caused by single monogenically inherited alleles but instead may be due to the cumulative influence of many genomic components. Although these components may have small individual effects on PD risk, they may have combined effect on a biological network that substantially influences PD risk. Consistent genome-wide patterns associated with risk for PD have been found for DNA modifications and SNPs. In order to identify biological processes involved in sporadic PD, we created gene networks based on human postmortem brain PD methylation profiles and human brain GWAS data, using a tool called NetWAS. This analysis uses integrated publicly available datasets to generate tissue-specific Bayesian approach to create tissue-specific interaction networks. Through this analysis we have identified potentially coordinated genetic and epigenetic networks, expanding our understanding of the genetic processes underlying disease. The genes that regulate these gene networks will be further studied in cell and animal models.

Short Abstract: Human primary tumors host a great diversity of mutations across the genome, preventing clear distinction between pathologically relevant driver and peripheral passenger mutations and complicating the study of biochemical reaction pathways and molecular mechanisms involved in cancer. Here, using high quality manually curated pathways annotated in Reactome, the most comprehensive open source biological pathway knowledgebase, along with the Mechsimo framework, a protein 3D structure-based approach to study contribution of amino acid changes to functions, we apply protein-protein and protein-chemical binding interface structural analysis to primary tumor samples from The Cancer Genome Atlas (TCGA) and show patient survival for some cancer types associates with functional interaction and biochemical reaction network features. Despite the relative paucity of publicly available experimentally derived protein structures and our incomplete knowledge of biochemical reaction pathways, focusing on mechanistic impact prediction of nonsynonymous mutations localized within protein interaction interfaces allows higher-order patterns to emerge exclusively from mutations likely affecting known cellular processes. These results provide insights potentiating 1) increased precision of cancer subtype characterization, 2) discovery and annotation of novel "long tail" mutations driving cancer, and 3) prediction and comprehension of response to personalized treatment.

Short Abstract: One of the major goals of systems biology is to de novo construct signaling pathways and generate testable hypotheses based on multi-omics datasets. Here we studied naïve T cell activation and performed temporal profiling of whole proteome (> 8,000 proteins) and phosphoproteome (> 13,000 phosphopeptides) via multiplexed isobaric labeling proteomics technology. Co-expression network analysis on both whole and phosphor-proteome defined protein and phosphopeptide clusters with distinct expression dynamics upon T cell stimulation, and the activity of upstream transcription factors (TFs) and kinases were inferred based on target expression and substrate phosphorylation level, respectively. To stratify downstream activated pathways, co-expression clusters were superimposed onto protein-protein interaction (PPI) network, identifying activated bioenergetics and mitochondrial pathways including mitoribosomes and complex IV. Finally, a signaling cascade was constructed by integrating kinase and TF activity, the kinase-substrate and TF-target circuits, as well as co-expression and PPI networks. This de novo assembled signaling pathway generates testable hypotheses, of which by genetic perturbation we confirmed the physiological roles of complex IV underlying T cell activation. Our result exemplifies the power of deep proteomics and the integration of multi-omics data in systems biology.

Short Abstract: The goal of this paper is to develop a novel statistical framework for inferring depedence between distributions of variables in omics data called ISaac. We propose the concept of building a dependence network using a copula-based kernel dependency measures to reconstruct the underlying association network between the distributions. The main idea is to combine empirical copula transformations with reproducing kernel based divergence estimators through Mean discrepancy measure. We prove that the estimator is consistent, robust to outliers, uses rank statistics only therefore is simple to derive therefore we propose two tests statistics to study the inferences: one based on convergence bounds and the second on the asymptotic distribution of the U-Statistic. Both are easy to implement. We noticed that by using the empirical copula transformation, the convergence rate was slightly affected, but the resulting dependence estimator possesses all the required properties for consistency. Results: Isaac is utilized for reverse-engineering gene regulatory networks and is competitive with several state-of-the-art gene regulatory inferrence methods on DREAM3 and DREAM4 Challenge datasets and on real data sets. Availability: An open-source implementation of ISaaC is available at \url{https://bitbucket.org/HossamAlmeer/isaac/

Short Abstract: In our two-level analysis, we first observe the capability of differential network analysis between normal and diseased state in detecting the important markers for cancer. We then, through a comparative study, propose isoform based differential network as a better alternative than its gene counter in identifying cancer biomarkers. To test the proposed methodology, two comparative studies have been done using data from Gene Expression Omnibus (GEO) and The Cancer Genome Atlas (TCGA). After preprocessing, we analyzed the data using Partial Least Square based Differential Network Analysis algorithm on most variable genes and isoforms. Differential network analysis resulted in each gene/isoform in the network with a connectivity score. Permutation tests based on these scores determined if the connectivity of a gene differs between the two networks. Top differentially connected genes/isoforms with p-value less than 0.05 were selected and Ingenuity Pathway Analysis (IPA) and Gene Set Enrichment Analysis is performed to find out the involvement of selected genes/isoforms in various pathways and GSEA genesets. . The methodology performed satisfactorily well in finding such markers genes and isoforms that are part of cancer related pathways. In addition, isoform level analysis showed better performance than the gene level study.