Advances to Bayesian network inference for generating causal networks from observational biological data

MOTIVATION: Network inference algorithms are powerful computational tools for identifying putative causal interactions among variables from observational data. Bayesian network inference algorithms hold particular promise in that they can capture linear, non-linear, combinatorial, stochastic and other types of relationships among variables across multiple levels of biological organization. However, challenges remain when applying these algorithms to limited quantities of experimental data collected from biological systems. Here, we use a simulation approach to make advances in our dynamic Bayesian network (DBN) inference algorithm, especially in the context of limited quantities of biological data. RESULTS: We test a range of scoring metrics and search heuristics to find an effective algorithm configuration for evaluating our methodological advances. We also identify sampling intervals and levels of data discretization that allow the best recovery of the simulated networks. We develop a novel influence score for DBNs that attempts to estimate both the sign (activation or repression) and relative magnitude of interactions among variables. When faced with limited quantities of observational data, combining our influence score with moderate data interpolation reduces a significant portion of false positive interactions in the recovered networks. Together, our advances allow DBN inference algorithms to be more effective in recovering biological networks from experimentally collected data. AVAILABILITY: Source code and simulated data are available upon request. SUPPLEMENTARY INFORMATION: http://www.jarvislab.net/Bioinformatics/BNAdvances/     

It's the machine that matters: Predicting gene function and phenotype from protein networks

Increasing knowledge about the organization of proteins into complexes, systems, and pathways has led to a flowering of theoretical approaches for exploiting this knowledge in order to better learn the functions of proteins and their roles underlying phenotypic traits and diseases. Much of this body of theory has been developed and tested in model organisms, relying on their relative simplicity and genetic and biochemical tractability to accelerate the research. In this review, we discuss several of the major approaches for computationally integrating proteomics and genomics observations into integrated protein networks, then applying guilt-by-association in these networks in order to identify genes underlying traits. Recent trends in this field include a rising appreciation of the modular network organization of proteins underlying traits or mutational phenotypes, and how to exploit such protein modularity using computational approaches related to the internet search algorithm PageRank. Many protein network-based predictions have recently been experimentally confirmed in yeast, worms, plants, and mice, and several successful approaches in model organisms have been directly translated to analyze human disease, with notable recent applications to glioma and breast cancer prognosis.     

Models for navigating biological complexity in breeding improved crop plants

Progress in breeding higher-yielding crop plants would be greatly accelerated if the phenotypic consequences of making changes to the genetic makeup of an organism could be reliably predicted. Developing a predictive capacity that scales from genotype to phenotype is impeded by biological complexities associated with genetic controls, environmental effects and interactions among plant growth and development processes. Plant modelling can help navigate a path through this complexity. Here we profile modelling approaches for complex traits at gene network, organ and whole plant levels. Each provides a means to link phenotypic consequence to changes in genomic regions via stable associations with model coefficients. A unifying feature of the models is the relatively coarse level of granularity they use to capture system dynamics. Much of the fine detail is not directly required. Robust coarse-grained models might be the tool needed to integrate phenotypic and molecular approaches to plant breeding.     

Structure‐energy‐based predictions and network modelling of RASopathy and cancer missense mutations

The Ras/MAPK syndromes (‘RASopathies’) are a class of developmental disorders caused by germline mutations in 15 genes encoding proteins of the Ras/mitogen‐activated protein kinase (MAPK) pathway frequently involved in cancer. Little is known about the molecular mechanisms underlying the differences in mutations of the same protein causing either cancer or RASopathies. Here, we shed light on 956 RASopathy and cancer missense mutations by combining protein network data with mutational analyses based on 3D structures. Using the protein design algorithm FoldX, we predict that most of the missense mutations with destabilising energies are in structural regions that control the activation of proteins, and only a few are predicted to compromise protein folding. We find a trend that energy changes are higher for cancer compared to RASopathy mutations. Through network modelling, we show that partly compensatory mutations in RASopathies result in only minor downstream pathway deregulation. In summary, we suggest that quantitative rather than qualitative network differences determine the phenotypic outcome of RASopathy compared to cancer mutations.     

Detecting and Removing Inconsistencies between Experimental Data and Signaling Network Topologies Using Integer Linear Programming on Interaction Graphs

Cross-referencing experimental data with our current knowledge of signaling network topologies is one central goal of mathematical modeling of cellular signal transduction networks. We present a new methodology for data-driven interrogation and training of signaling networks. While most published methods for signaling network inference operate on Bayesian, Boolean, or ODE models, our approach uses integer linear programming (ILP) on interaction graphs to encode constraints on the qualitative behavior of the nodes. These constraints are posed by the network topology and their formulation as ILP allows us to predict the possible qualitative changes (up, down, no effect) of the activation levels of the nodes for a given stimulus. We provide four basic operations to detect and remove inconsistencies between measurements and predicted behavior: (i) find a topology-consistent explanation for responses of signaling nodes measured in a stimulus-response experiment (if none exists, find the closest explanation); (ii) determine a minimal set of nodes that need to be corrected to make an inconsistent scenario consistent; (iii) determine the optimal subgraph of the given network topology which can best reflect measurements from a set of experimental scenarios; (iv) find possibly missing edges that would improve the consistency of the graph with respect to a set of experimental scenarios the most. We demonstrate the applicability of the proposed approach by interrogating a manually curated interaction graph model of EGFR/ErbB signaling against a library of high-throughput phosphoproteomic data measured in primary hepatocytes. Our methods detect interactions that are likely to be inactive in hepatocytes and provide suggestions for new interactions that, if included, would significantly improve the goodness of fit. Our framework is highly flexible and the underlying model requires only easily accessible biological knowledge. All related algorithms were implemented in a freely available toolbox SigNetTrainer making it an appealing approach for various applications.     

Model of interactions in biology and application to heterogeneous network in yeast

A major challenge for bioinformatics and theoretical biology is to build and analyse a unified model of biological knowledge resulting from high-throughput experiment data. Former work analyzed heterogeneous data (protein-protein interactions, genetic regulation, metabolism, synexpression) by modelling them by graphs. These models are unable to represent the qualitative dynamics of the reactions or to model the n-ary interactions. Here, MIB, the Model of Interactions in Biology, a bipartite model of biological networks, is introduced, and its use for topological analysis of the heterogeneous network is presented. Heterogeneous loops and links between synexpression pattern and underlying molecular mechanisms are proposed.     

Epistasis in a quantitative trait captured by a molecular model of transcription factor interactions

With technological advances in genetic mapping studies more of the genes and polymorphisms that underlie Quantitative Trait Loci (QTL) are now being identified. As the identities of these genes become known there is a growing need for an analysis framework that incorporates the molecular interactions affected by natural polymorphisms. As a step towards such a framework we present a molecular model of genetic variation in sporulation efficiency between natural isolates of the yeast, Saccharomyces cerevisiae. The model is based on the structure of the regulatory pathway that controls sporulation. The model captures the phenotypic variation between strains carrying different combinations of alleles at known QTL. Compared to a standard linear model the molecular model requires fewer free parameters, and has the advantage of generating quantitative hypotheses about the affinity of specific molecular interactions in different genetic backgrounds. Our analyses provide a concrete example of how the thermodynamic properties of protein-protein and protein-DNA interactions naturally give rise to epistasis, the non-linear relationship between genotype and phenotype. As more causative genes and polymorphisms underlying QTL are identified, thermodynamic analyses of quantitative traits may provide a useful framework for unraveling the complex relationship between genotype and phenotype.     

Developmental interactions and the constituents of quantitative variation

Development is the process by which genotypes are transformed into phenotypes. Consequently, development determines the relationship between allelic and phenotypic variation in a population and, therefore, the patterns of quantitative genetic variation and covariation of traits. Understanding the developmental basis of quantitative traits may lead to insights into the origin and evolution of quantitative genetic variation, the evolutionary fate of populations, and, more generally, the relationship between development and evolution. Herein, we assume a hierarchical, modular structure of trait development and consider how epigenetic interactions among modules during ontogeny affect patterns of phenotypic and genetic variation. We explore two developmental models, one in which the epigenetic interactions between modules result in additive effects on character expression and a second model in which these epigenetic interactions produce nonadditive effects. Using a phenotype landscape approach, we show how changes in the developmental processes underlying phenotypic expression can alter the magnitude and pattern of quantitative genetic variation. Additive epigenetic effects influence genetic variances and covariances, but allow trait means to evolve independently of the genetic variances and covariances, so that phenotypic evolution can proceed without changing the genetic covariance structure that determines future evolutionary response. Nonadditive epigenetic effects, however, can lead to evolution of genetic variances and covariances as the mean phenotype evolves. Our model suggests that an understanding of multivariate evolution can be considerably enriched by knowledge of the mechanistic basis of character development.     

Genomic characterization of multiple clinical phenotypes of cancer using multivariate linear regression models

MOTIVATION: The development of gene expression microarray technology has allowed the identification of differentially expressed genes between different clinical phenotypic classes of cancer from a large pool of candidate genes. Although many class comparisons concerned only a single phenotype, simultaneous assessment of the relationship between gene expression and multiple phenotypes would be warranted to better understand the underlying biological structure. RESULTS: We develop a method to select genes related to multiple clinical phenotypes based on a set of multivariate linear regression models. For each gene, we perform model selection based on the doubly-adjusted R-square statistic and use the maximum of this statistic for gene selection. The method can substantially improve the power in gene selection, compared with a conventional method that uses a single model exclusively for gene selection. Application to a bladder cancer study to correlate pre-treatment gene expressions with pathological stage and grade is given. The methods would be useful for screening for genes related to multiple clinical phenotypes. AVAILABILITY: SAS and MATLAB codes are available from author upon request.     

Inferring network interactions within a cell

The continuing growth in high-throughput data acquisition has led to a proliferation of network models to represent and analyse biological systems. These networks involve distinct interaction types detected by a combination of methods, ranging from directly observed physical interactions based in biochemistry to interactions inferred from phenotype measurements, genomic expression and comparative genomics. The discovery of interactions increasingly requires a blend of experimental and computational methods. Considering yeast as a model system, recent analytical methods are reviewed here and specific aims are proposed to improve network interaction inference and facilitate predictive biological modelling.     

Evolutionary conservation and network structure characterize genes of phenotypic relevance for mitosis in human

The impact of gene silencing on cellular phenotypes is difficult to establish due to the complexity of interactions in the associated biological processes and pathways. A recent genome-wide RNA knock-down study both identified and phenotypically characterized a set of important genes for the cell cycle in HeLa cells. Here, we combine a molecular interaction network analysis, based on physical and functional protein interactions, in conjunction with evolutionary information, to elucidate the common biological and topological properties of these key genes. Our results show that these genes tend to be conserved with their corresponding protein interactions across several species and are key constituents of the evolutionary conserved molecular interaction network. Moreover, a group of bistable network motifs is found to be conserved within this network, which are likely to influence the network stability and therefore the robustness of cellular functioning. They form a cluster, which displays functional homogeneity and is significantly enriched in genes phenotypically relevant for mitosis. Additional results reveal a relationship between specific cellular processes and the phenotypic outcomes induced by gene silencing. This study introduces new ideas regarding the relationship between genotype and phenotype in the context of the cell cycle. We show that the analysis of molecular interaction networks can result in the identification of genes relevant to cellular processes, which is a promising avenue for future research.     

Revealing Molecular Mechanisms by Integrating High-Dimensional Functional Screens with Protein Interaction Data

Functional genomics screens using multi-parametric assays are powerful approaches for identifying genes involved in particular cellular processes. However, they suffer from problems like noise, and often provide little insight into molecular mechanisms. A bottleneck for addressing these issues is the lack of computational methods for the systematic integration of multi-parametric phenotypic datasets with molecular interactions. Here, we present Integrative Multi Profile Analysis of Cellular Traits (IMPACT). The main goal of IMPACT is to identify the most consistent phenotypic profile among interacting genes. This approach utilizes two types of external information: sets of related genes (IMPACT-sets) and network information (IMPACT-modules). Based on the notion that interacting genes are more likely to be involved in similar functions than non-interacting genes, this data is used as a prior to inform the filtering of phenotypic profiles that are similar among interacting genes. IMPACT-sets selects the most frequent profile among a set of related genes. IMPACT-modules identifies sub-networks containing genes with similar phenotype profiles. The statistical significance of these selections is subsequently quantified via permutations of the data. IMPACT (1) handles multiple profiles per gene, (2) rescues genes with weak phenotypes and (3) accounts for multiple biases e.g. caused by the network topology. Application to a genome-wide RNAi screen on endocytosis showed that IMPACT improved the recovery of known endocytosis-related genes, decreased off-target effects, and detected consistent phenotypes. Those findings were confirmed by rescreening 468 genes. Additionally we validated an unexpected influence of the IGF-receptor on EGF-endocytosis. IMPACT facilitates the selection of high-quality phenotypic profiles using different types of independent information, thereby supporting the molecular interpretation of functional screens.     

Decipher the connections between proteins and phenotypes

As the outward-most representation of life, phenotype is the fundamental basis with which humans understand life and disease. But with the advent of molecular and sequencing technique and research, a growing portion of science research focuses primarily on the molecular level of life. Our understanding in molecular variations and mechanisms can only be fully utilized when they are translated into the phenotypic level. In this study, we constructed similarity network for phenotype ontology, and then applied network analysis methods to discover phenotype/disease clusters. Then, we used machine learning models to predict protein-phenotype associations. Each protein was characterized by the functional profiles of its interaction neighbors on the protein-protein interaction network. Our methods can not only predict protein-phenotype associations, but also reveal the underlying mechanisms from protein to phenotype.     

A protein interaction network associated with asthma

Identifying candidate genes related to complex diseases or traits and mapping their relationships require a system-level analysis at a cellular scale. The objective of the present study is to systematically analyze the complex effects of interrelated genes and provide a framework for revealing their relationships in association with a specific disease (asthma in this case). We observed that protein-protein interaction (PPI) networks associated with asthma have a power-law connectivity distribution as many other biological networks have. The hub nodes and skeleton substructure of the result network are consistent with the prior knowledge about asthma pathways, and also suggest unknown candidate target genes associated with asthma, including GNB2L1, BRCA1, CBL, and VAV1. In particular, GNB2L1 appears to play a very important role in the asthma network through frequent interactions with key proteins in cellular signaling. This network-based approach represents an alternative method for analyzing the complex effects of candidate genes associated with complex diseases and suggesting a list of gene drug targets. The full list of genes and the analysis details are available in the following online supplementary materials: http://biosoft.kaist.ac.kr:8080/resources/asthma_ppi.     

Molecular network strategy in multi-omics and mass spectrometry imaging

Human physiological activities and pathological changes arise from the coordinated interactions of multiple molecules. Mass spectrometry (MS)-based multi-omics and MS imaging (MSI)-based spatial omics are powerful methods used to investigate molecular information related to the phenotype of interest from homogenated or sliced samples, including the qualitative, relative quantitative and spatial distributions. Molecular network strategy provides efficient methods to help us understand and mine the biological patterns behind the phenotypic data. It illustrates and combines various relationships between molecules, and further performs the molecule identification and biological interpretation. Here, we describe the recent advances of network-based analysis and its applications for different biological processes, such as, obesity, central nervous system diseases, and environmental toxicology.     

Computational identification of protein-protein interactions in rice based on the predicted rice interactome network

Plant protein-protein interaction networks have not been identified by large-scale experiments. In order to better understand the protein interactions in rice, the Predicted Rice Interactome Network (PRIN; http://bis.zju.edu.cn/prin/) presented 76,585 predicted interactions involving 5,049 rice proteins. After mapping genomic features of rice (GO annotation, subcellular localization prediction, and gene expression), we found that a well-annotated and biologically significant network is rich enough to capture many significant functional linkages within higher-order biological systems, such as pathways and biological processes. Furthermore, we took MADS-box domain-containing proteins and circadian rhythm signaling pathways as examples to demonstrate that functional protein complexes and biological pathways could be effectively expanded in our predicted network. The expanded molecular network in PRIN has considerably improved the capability of these analyses to integrate existing knowledge and provide novel insights into the function and coordination of genes and gene networks.     

Gene Networks Underlying Convergent and Pleiotropic Phenotypes in a Large and Systematically-Phenotyped Cohort with Heterogeneous Developmental Disorders

Readily-accessible and standardised capture of genotypic variation has revolutionised our understanding of the genetic contribution to disease. Unfortunately, the corresponding systematic capture of patient phenotypic variation needed to fully interpret the impact of genetic variation has lagged far behind. Exploiting deep and systematic phenotyping of a cohort of 197 patients presenting with heterogeneous developmental disorders and whose genomes harbour de novo CNVs, we systematically applied a range of commonly-used functional genomics approaches to identify the underlying molecular perturbations and their phenotypic impact. Grouping patients into 408 non-exclusive patient-phenotype groups, we identified a functional association amongst the genes disrupted in 209 (51%) groups. We find evidence for a significant number of molecular interactions amongst the association-contributing genes, including a single highly-interconnected network disrupted in 20% of patients with intellectual disability, and show using microcephaly how these molecular networks can be used as baits to identify additional members whose genes are variant in other patients with the same phenotype. Exploiting the systematic phenotyping of this cohort, we observe phenotypic concordance amongst patients whose variant genes contribute to the same functional association but note that (i) this relationship shows significant variation across the different approaches used to infer a commonly perturbed molecular pathway, and (ii) that the phenotypic similarities detected amongst patients who share the same inferred pathway perturbation result from these patients sharing many distinct phenotypes, rather than sharing a more specific phenotype, inferring that these pathways are best characterized by their pleiotropic effects.