Inference and validation of predictive gene networks from biomedical literature and gene expression data

Predicting transcription factor activity using prior biological information

Dysregulation of normal transcription factor activity is a common driver of disease. Therefore, the detection of aberrant transcription factor activity is important to understand disease pathogenesis. We have developed Priori, a method to predict transcription factor activity from RNA sequencing data. Priori has two key advantages over existing methods. First, Priori utilizes literature-supported regulatory information to identify transcription factor-target gene relationships. It then applies linear models to determine the impact of transcription factor regulation on the expression of its target genes. Second, results from a third-party benchmarking pipeline reveals that Priori detects aberrant activity from 124 single-gene perturbation experiments with higher sensitivity and specificity than 11 other methods. We applied Priori and other top-performing methods to predict transcription factor activity from two large primary patient datasets. Our work demonstrates that Priori uniquely discovered significant determinants of survival in breast cancer and identified mediators of drug response in leukemia.

A large sequenced mutant library - valuable reverse genetic resource that covers 98% of sorghum genes

Mutant populations are crucial for functional genomics and discovering novel traits for crop breeding. Sorghum, a drought and heat-tolerant C4 species, requires a vast, large-scale, annotated, and sequenced mutant resource to enhance crop improvement through functional genomics research. Here, we report a sorghum large-scale sequenced mutant population with 9.5 million ethyl methane sulfonate (EMS)-induced mutations that covered 98% of sorghum's annotated genes using inbred line BTx623. Remarkably, a total of 610 320 mutations within the promoter and enhancer regions of 18 000 and 11 790 genes, respectively, can be leveraged for novel research of cis-regulatory elements. A comparison of the distribution of mutations in the large-scale mutant library and sorghum association panel (SAP) provides insights into the influence of selection. EMS-induced mutations appeared to be random across different regions of the genome without significant enrichment in different sections of a gene, including the 5' UTR, gene body, and 3'-UTR. In contrast, there were low variation density in the coding and UTR regions in the SAP. Based on the Ka /Ks value, the mutant library (~1) experienced little selection, unlike the SAP (0.40), which has been strongly selected through breeding. All mutation data are publicly searchable through SorbMutDB (https://www.depts.ttu.edu/igcast/sorbmutdb.php) and SorghumBase (https://sorghumbase.org/). This current large-scale sequence-indexed sorghum mutant population is a crucial resource that enriched the sorghum gene pool with novel diversity and a highly valuable tool for the Poaceae family, that will advance plant biology research and crop breeding.

RNA-seq preprocessing and sample size considerations for gene network inference

Background

Gene network inference (GNI) methods have the potential to reveal functional relationships between different genes and their products. Most GNI algorithms have been developed for microarray gene expression datasets and their application to RNA-seq data is relatively recent. As the characteristics of RNA-seq data are different from microarray data, it is an unanswered question what preprocessing methods for RNA-seq data should be applied prior to GNI to attain optimal performance, or what the required sample size for RNA-seq data is to obtain reliable GNI estimates.

Results

We ran 9144 analysis of 7 different RNA-seq datasets to evaluate 300 different preprocessing combinations that include data transformations, normalizations and association estimators. We found that there was no single best performing preprocessing combination but that there were several good ones. The performance varied widely over various datasets, which emphasized the importance of choosing an appropriate preprocessing configuration before GNI. Two preprocessing combinations appeared promising in general: First, Log-2 TPM (transcript per million) with Variance-stabilizing transformation (VST) and Pearson Correlation Coefficient (PCC) association estimator. Second, raw RNA-seq count data with PCC. Along with these two, we also identified 18 other good preprocessing combinations. Any of these algorithms might perform best in different datasets. Therefore, the GNI performances of these approaches should be measured on any new dataset to select the best performing one for it. In terms of the required biological sample size of RNA-seq data, we found that between 30 to 85 samples were required to generate reliable GNI estimates.

Conclusions

This study provides practical recommendations on default choices for data preprocessing prior to GNI analysis of RNA-seq data to obtain optimal performance results.

WINNER: A network biology tool for biomolecular characterization and prioritization

BACKGROUND AND CONTRIBUTION

In network biology, molecular functions can be characterized by network-based inference, or "guilt-by-associations." PageRank-like tools have been applied in the study of biomolecular interaction networks to obtain further the relative significance of all molecules in the network. However, there is a great deal of inherent noise in widely accessible data sets for gene-to-gene associations or protein-protein interactions. How to develop robust tests to expand, filter, and rank molecular entities in disease-specific networks remains an ad hoc data analysis process.

RESULTS

We describe a new biomolecular characterization and prioritization tool called Weighted In-Network Node Expansion and Ranking (WINNER). It takes the input of any molecular interaction network data and generates an optionally expanded network with all the nodes ranked according to their relevance to one another in the network. To help users assess the robustness of results, WINNER provides two different types of statistics. The first type is a node-expansion p-value, which helps evaluate the statistical significance of adding "non-seed" molecules to the original biomolecular interaction network consisting of "seed" molecules and molecular interactions. The second type is a node-ranking p-value, which helps evaluate the relative statistical significance of the contribution of each node to the overall network architecture. We validated the robustness of WINNER in ranking top molecules by spiking noises in several network permutation experiments. We have found that node degree-preservation randomization of the gene network produced normally distributed ranking scores, which outperform those made with other gene network randomization techniques. Furthermore, we validated that a more significant proportion of the WINNER-ranked genes was associated with disease biology than existing methods such as PageRank. We demonstrated the performance of WINNER with a few case studies, including Alzheimer's disease, breast cancer, myocardial infarctions, and Triple negative breast cancer (TNBC). In all these case studies, the expanded and top-ranked genes identified by WINNER reveal disease biology more significantly than those identified by other gene prioritizing software tools, including Ingenuity Pathway Analysis (IPA) and DiAMOND.

CONCLUSION

WINNER ranking strongly correlates to other ranking methods when the network covers sufficient node and edge information, indicating a high network quality. WINNER users can use this new tool to robustly evaluate a list of candidate genes, proteins, or metabolites produced from high-throughput biology experiments, as long as there is available gene/protein/metabolic network information.

Perturbation-based gene regulatory network inference to unravel oncogenic mechanisms

The gene regulatory network (GRN) of human cells encodes mechanisms to ensure proper functioning. However, if this GRN is dysregulated, the cell may enter into a disease state such as cancer. Understanding the GRN as a system can therefore help identify novel mechanisms underlying disease, which can lead to new therapies. To deduce regulatory interactions relevant to cancer, we applied a recent computational inference framework to data from perturbation experiments in squamous carcinoma cell line A431. GRNs were inferred using several methods, and the false discovery rate was controlled by the NestBoot framework. We developed a novel approach to assess the predictiveness of inferred GRNs against validation data, despite the lack of a gold standard. The best GRN was significantly more predictive than the null model, both in cross-validated benchmarks and for an independent dataset of the same genes under a different perturbation design. The inferred GRN captures many known regulatory interactions central to cancer-relevant processes in addition to predicting many novel interactions, some of which were experimentally validated, thus providing mechanistic insights that are useful for future cancer research.

Inference of Genome-Scale Gene Regulatory Networks: Are There Differences in Biological and Clinical Validations?

Causal networks, e.g., gene regulatory networks (GRNs) inferred from gene expression data, contain a wealth of information but are defying simple, straightforward and low-budget experimental validations. In this paper, we elaborate on this problem and discuss distinctions between biological and clinical validations. As a result, validation differences for GRNs reflect known differences between basic biological and clinical research questions making the validations context specific. Hence, the meaning of biologically and clinically meaningful GRNs can be very different. For a concerted approach to a problem of this size, we suggest the establishment of the HUMAN GENE REGULATORY NETWORK PROJECT which provides the information required for biological and clinical validations alike.

RGBM: regularized gradient boosting machines for identification of the transcriptional regulators of discrete glioma subtypes

We propose a generic framework for gene regulatory network (GRN) inference approached as a feature selection problem. GRNs obtained using Machine Learning techniques are often dense, whereas real GRNs are rather sparse. We use a Tikonov regularization inspired optimal L-curve criterion that utilizes the edge weight distribution for a given target gene to determine the optimal set of TFs associated with it. Our proposed framework allows to incorporate a mechanistic active biding network based on cis-regulatory motif analysis. We evaluate our regularization framework in conjunction with two non-linear ML techniques, namely gradient boosting machines (GBM) and random-forests (GENIE), resulting in a regularized feature selection based method specifically called RGBM and RGENIE respectively. RGBM has been used to identify the main transcription factors that are causally involved as master regulators of the gene expression signature activated in the FGFR3-TACC3-positive glioblastoma. Here, we illustrate that RGBM identifies the main regulators of the molecular subtypes of brain tumors. Our analysis reveals the identity and corresponding biological activities of the master regulators characterizing the difference between G-CIMP-high and G-CIMP-low subtypes and between PA-like and LGm6-GBM, thus providing a clue to the yet undetermined nature of the transcriptional events among these subtypes.

Analytical Strategy to Prioritize Alzheimer's Disease Candidate Genes in Gene Regulatory Networks Using Public Expression Data

Alzheimer’s disease (AD) progressively destroys cognitive abilities in the aging population with tremendous effects on memory. Despite recent progress in understanding the underlying mechanisms, high drug attrition rates have put a question mark behind our knowledge about its etiology. Re-evaluation of past studies could help us to elucidate molecular-level details of this disease. Several methods to infer such networks exist, but most of them do not elaborate on context specificity and completeness of the generated networks, missing out on lesser-known candidates. In this study, we present a novel strategy that corroborates common mechanistic patterns across large scale AD gene expression studies and further prioritizes potential biomarker candidates. To infer gene regulatory networks (GRNs), we applied an optimized version of the BC3Net algorithm, named BC3Net10, capable of deriving robust and coherent patterns. In principle, this approach initially leverages the power of literature knowledge to extract AD specific genes for generating viable networks. Our findings suggest that AD GRNs show significant enrichment for key signaling mechanisms involved in neurotransmission. Among the prioritized genes, well-known AD genes were prominent in synaptic transmission, implicated in cognitive deficits. Moreover, less intensive studied AD candidates (STX2, HLA-F, HLA-C, RAB11FIP4, ARAP3, AP2A2, ATP2B4, ITPR2, and ATP2A3) are also involved in neurotransmission, providing new insights into the underlying mechanism. To our knowledge, this is the first study to generate knowledge-instructed GRNs that demonstrates an effective way of combining literature-based knowledge and data-driven analysis to identify lesser known candidates embedded in stable and robust functional patterns across disparate datasets.

Construction and analysis of gene-gene dynamics influence networks based on a Boolean model

BACKGROUND

Identification of novel gene-gene relations is a crucial issue to understand system-level biological phenomena. To this end, many methods based on a correlation analysis of gene expressions or structural analysis of molecular interaction networks have been proposed. They have a limitation in identifying more complicated gene-gene dynamical relations, though.

RESULTS

To overcome this limitation, we proposed a measure to quantify a gene-gene dynamical influence (GDI) using a Boolean network model and constructed a GDI network to indicate existence of a dynamical influence for every ordered pair of genes. It represents how much a state trajectory of a target gene is changed by a knockout mutation subject to a source gene in a gene-gene molecular interaction (GMI) network. Through a topological comparison between GDI and GMI networks, we observed that the former network is denser than the latter network, which implies that there exist many gene pairs of dynamically influencing but molecularly non-interacting relations. In addition, a larger number of hub genes were generated in the GDI network. On the other hand, there was a correlation between these networks such that the degree value of a node was positively correlated to each other. We further investigated the relationships of the GDI value with structural properties and found that there are negative and positive correlations with the length of a shortest path and the number of paths, respectively. In addition, a GDI network could predict a set of genes whose steady-state expression is affected in E. coli gene-knockout experiments. More interestingly, we found that the drug-targets with side-effects have a larger number of outgoing links than the other genes in the GDI network, which implies that they are more likely to influence the dynamics of other genes. Finally, we found biological evidences showing that the gene pairs which are not molecularly interacting but dynamically influential can be considered for novel gene-gene relationships.

CONCLUSION

Taken together, construction and analysis of the GDI network can be a useful approach to identify novel gene-gene relationships in terms of the dynamical influence.

GFD-Net: A novel semantic similarity methodology for the analysis of gene networks

Since the popularization of biological network inference methods, it has become crucial to create methods to validate the resulting models. Here we present GFD-Net, the first methodology that applies the concept of semantic similarity to gene network analysis. GFD-Net combines the concept of semantic similarity with the use of gene network topology to analyze the functional dissimilarity of gene networks based on Gene Ontology (GO). The main innovation of GFD-Net lies in the way that semantic similarity is used to analyze gene networks taking into account the network topology. GFD-Net selects a functionality for each gene (specified by a GO term), weights each edge according to the dissimilarity between the nodes at its ends and calculates a quantitative measure of the network functional dissimilarity, i.e. a quantitative value of the degree of dissimilarity between the connected genes. The robustness of GFD-Net as a gene network validation tool was demonstrated by performing a ROC analysis on several network repositories. Furthermore, a well-known network was analyzed showing that GFD-Net can also be used to infer knowledge. The relevance of GFD-Net becomes more evident in Section "GFD-Net applied to the study of human diseases" where an example of how GFD-Net can be applied to the study of human diseases is presented. GFD-Net is available as an open-source Cytoscape app which offers a user-friendly interface to configure and execute the algorithm as well as the ability to visualize and interact with the results(http://apps.cytoscape.org/apps/gfdnet).

Systems Biology Approaches for Host-Fungal Interactions: An Expanding Multi-Omics Frontier

Opportunistic fungal infections are an increasing threat for global health, and for immunocompromised patients in particular. These infections are characterized by interaction between fungal pathogen and host cells. The exact mechanisms and the attendant variability in host and fungal pathogen interaction remain to be fully elucidated. The field of systems biology aims to characterize a biological system, and utilize this knowledge to predict the system's response to stimuli such as fungal exposures. A multi-omics approach, for example, combining data from genomics, proteomics, metabolomics, would allow a more comprehensive and pan-optic "two systems" biology of both the host and the fungal pathogen. In this review and literature analysis, we present highly specialized and nascent methods for analysis of multiple -omes of biological systems, in addition to emerging single-molecule visualization techniques that may assist in determining biological relevance of multi-omics data. We provide an overview of computational methods for modeling of gene regulatory networks, including some that have been applied towards the study of an interacting host and pathogen. In sum, comprehensive characterizations of host-fungal pathogen systems are now possible, and utilization of these cutting-edge multi-omics strategies may yield advances in better understanding of both host biology and fungal pathogens at a systems scale.

Inferring causal molecular networks: empirical assessment through a community-based effort

The HPN-DREAM community challenge assessed the ability of computational methods to infer causal molecular networks, focusing specifically on the task of inferring causal protein signaling networks in cancer cell lines.

It remains unclear whether causal, rather than merely correlational, relationships in molecular networks can be inferred in complex biological settings. Here we describe the HPN-DREAM network inference challenge, which focused on learning causal influences in signaling networks. We used phosphoprotein data from cancer cell lines as well as in silico data from a nonlinear dynamical model. Using the phosphoprotein data, we scored more than 2,000 networks submitted by challenge participants. The networks spanned 32 biological contexts and were scored in terms of causal validity with respect to unseen interventional data. A number of approaches were effective, and incorporating known biology was generally advantageous. Additional sub-challenges considered time-course prediction and visualization. Our results suggest that learning causal relationships may be feasible in complex settings such as disease states. Furthermore, our scoring approach provides a practical way to empirically assess inferred molecular networks in a causal sense.

Taking Bioinformatics to Systems Medicine

Systems medicine promotes a range of approaches and strategies to study human health and disease at a systems level with the aim of improving the overall well-being of (healthy) individuals, and preventing, diagnosing, or curing disease. In this chapter we discuss how bioinformatics critically contributes to systems medicine. First, we explain the role of bioinformatics in the management and analysis of data. In particular we show the importance of publicly available biological and clinical repositories to support systems medicine studies. Second, we discuss how the integration and analysis of multiple types of omics data through integrative bioinformatics may facilitate the determination of more predictive and robust disease signatures, lead to a better understanding of (patho)physiological molecular mechanisms, and facilitate personalized medicine. Third, we focus on network analysis and discuss how gene networks can be constructed from omics data and how these networks can be decomposed into smaller modules. We discuss how the resulting modules can be used to generate experimentally testable hypotheses, provide insight into disease mechanisms, and lead to predictive models. Throughout, we provide several examples demonstrating how bioinformatics contributes to systems medicine and discuss future challenges in bioinformatics that need to be addressed to enable the advancement of systems medicine.

Using shRNA experiments to validate gene regulatory networks

Quantitative validation of gene regulatory networks (GRNs) inferred from observational expression data is a difficult task usually involving time intensive and costly laboratory experiments. We were able to show that gene knock-down experiments can be used to quantitatively assess the quality of large-scale GRNs via a purely data-driven approach (Olsen et al. 2014). Our new validation framework also enables the statistical comparison of multiple network inference techniques, which was a long-standing challenge in the field.

In this Data in Brief we detail the contents and quality controls for the gene expression data (available from NCBI Gene Expression Omnibus repository with accession number GSE53091) associated with our study published in Genomics (Olsen et al. 2014). We also provide R code to access the data and reproduce the analysis presented in this article.

Systems Genetics Analysis of Genome-Wide Association Study Reveals Novel Associations Between Key Biological Processes and Coronary Artery Disease

OBJECTIVE

Genome-wide association studies have identified multiple genetic variants affecting the risk of coronary artery disease (CAD). However, individually these explain only a small fraction of the heritability of CAD and for most, the causal biological mechanisms remain unclear. We sought to obtain further insights into potential causal processes of CAD by integrating large-scale GWA data with expertly curated databases of core human pathways and functional networks.

APPROACHES AND RESULTS

Using pathways (gene sets) from Reactome, we carried out a 2-stage gene set enrichment analysis strategy. From a meta-analyzed discovery cohort of 7 CAD genome-wide association study data sets (9889 cases/11 089 controls), nominally significant gene sets were tested for replication in a meta-analysis of 9 additional studies (15 502 cases/55 730 controls) from the Coronary ARtery DIsease Genome wide Replication and Meta-analysis (CARDIoGRAM) Consortium. A total of 32 of 639 Reactome pathways tested showed convincing association with CAD (replication P<0.05). These pathways resided in 9 of 21 core biological processes represented in Reactome, and included pathways relevant to extracellular matrix (ECM) integrity, innate immunity, axon guidance, and signaling by PDRF (platelet-derived growth factor), NOTCH, and the transforming growth factor-β/SMAD receptor complex. Many of these pathways had strengths of association comparable to those observed in lipid transport pathways. Network analysis of unique genes within the replicated pathways further revealed several interconnected functional and topologically interacting modules representing novel associations (eg, semaphoring-regulated axonal guidance pathway) besides confirming known processes (lipid metabolism). The connectivity in the observed networks was statistically significant compared with random networks (P<0.001). Network centrality analysis (degree and betweenness) further identified genes (eg, NCAM1, FYN, FURIN, etc) likely to play critical roles in the maintenance and functioning of several of the replicated pathways.

CONCLUSIONS

These findings provide novel insights into how genetic variation, interpreted in the context of biological processes and functional interactions among genes, may help define the genetic architecture of CAD.

How difficult is inference of mammalian causal gene regulatory networks?

Gene regulatory networks (GRNs) play a central role in systems biology, especially in the study of mammalian organ development. One key question remains largely unanswered: Is it possible to infer mammalian causal GRNs using observable gene co-expression patterns alone? We assembled two mouse GRN datasets (embryonic tooth and heart) and matching microarray gene expression profiles to systematically investigate the difficulties of mammalian causal GRN inference. The GRNs were assembled based on pieces of experimental genetic perturbation evidence from manually reading primary research articles. Each piece of perturbation evidence records the qualitative change of the expression of one gene following knock-down or over-expression of another gene. Our data have thorough annotation of tissue types and embryonic stages, as well as the type of regulation (activation, inhibition and no effect), which uniquely allows us to estimate both sensitivity and specificity of the inference of tissue specific causal GRN edges. Using these unprecedented datasets, we found that gene co-expression does not reliably distinguish true positive from false positive interactions, making inference of GRN in mammalian development very difficult. Nonetheless, if we have expression profiling data from genetic or molecular perturbation experiments, such as gene knock-out or signalling stimulation, it is possible to use the set of differentially expressed genes to recover causal regulatory relationships with good sensitivity and specificity. Our result supports the importance of using perturbation experimental data in causal network reconstruction. Furthermore, we showed that causal gene regulatory relationship can be highly cell type or developmental stage specific, suggesting the importance of employing expression profiles from homogeneous cell populations. This study provides essential datasets and empirical evidence to guide the development of new GRN inference methods for mammalian organ development.

Gene regulatory networks and their applications: understanding biological and medical problems in terms of networks

In recent years gene regulatory networks (GRNs) have attracted a lot of interest and many methods have been introduced for their statistical inference from gene expression data. However, despite their popularity, GRNs are widely misunderstood. For this reason, we provide in this paper a general discussion and perspective of gene regulatory networks. Specifically, we discuss their meaning, the consistency among different network inference methods, ensemble methods, the assessment of GRNs, the estimated number of existing GRNs and their usage in different application domains. Furthermore, we discuss open questions and necessary steps in order to utilize gene regulatory networks in a clinical context and for personalized medicine.

Relevance of different prior knowledge sources for inferring gene interaction networks

When inferring networks from high-throughput genomic data, one of the main challenges is the subsequent validation of these networks. In the best case scenario, the true network is partially known from previous research results published in structured databases or research articles. Traditionally, inferred networks are validated against these known interactions. Whenever the recovery rate is gauged to be high enough, subsequent high scoring but unknown inferred interactions are deemed good candidates for further experimental validation. Therefore such validation framework strongly depends on the quantity and quality of published interactions and presents serious pitfalls: (1) availability of these known interactions for the studied problem might be sparse; (2) quantitatively comparing different inference algorithms is not trivial; and (3) the use of these known interactions for validation prevents their integration in the inference procedure. The latter is particularly relevant as it has recently been showed that integration of priors during network inference significantly improves the quality of inferred networks. To overcome these problems when validating inferred networks, we recently proposed a data-driven validation framework based on single gene knock-down experiments. Using this framework, we were able to demonstrate the benefits of integrating prior knowledge and expression data. In this paper we used this framework to assess the quality of different sources of prior knowledge on their own and in combination with different genomic data sets in colorectal cancer. We observed that most prior sources lead to significant F-scores. Furthermore, their integration with genomic data leads to a significant increase in F-scores, especially for priors extracted from full text PubMed articles, known co-expression modules and genetic interactions. Lastly, we observed that the results are consistent for three different data sets: experimental knock-down data and two human tumor data sets.