Differential network analysis reveals the genome-wide landscape of estrogen receptor modulation in hormonal cancers

Interpretable AI for inference of causal molecular relationships from omics data

The discovery of molecular relationships from high-dimensional data is a major open problem in bioinformatics. Machine learning and feature attribution models have shown great promise in this context but lack causal interpretation. Here, we show that a popular feature attribution model, under certain assumptions, estimates an average of a causal quantity reflecting the direct influence of one variable on another. We leverage this insight to propose a precise definition of a gene regulatory relationship and implement a new tool, CIMLA (Counterfactual Inference by Machine Learning and Attribution Models), to identify differences in gene regulatory networks between biological conditions, a problem that has received great attention in recent years. Using extensive benchmarking on simulated data, we show that CIMLA is more robust to confounding variables and is more accurate than leading methods. Last, we use CIMLA to analyze a previously published single-cell RNA sequencing dataset from subjects with and without Alzheimer's disease (AD), discovering several potential regulators of AD.

Patient-specific analysis of co-expression to measure biological network rewiring in individuals

To effectively understand the underlying mechanisms of disease and inform the development of personalized therapies, it is critical to harness the power of differential co-expression (DCE) network analysis. Despite the promise of DCE network analysis in precision medicine, current approaches have a major limitation: they measure an average differential network across multiple samples, which means the specific etiology of individual patients is often overlooked. To address this, we present Cosinet, a DCE-based single-sample network rewiring degree quantification tool. By analyzing two breast cancer datasets, we demonstrate that Cosinet can identify important differences in gene co-expression patterns between individual patients and generate scores for each individual that are significantly associated with overall survival, recurrence-free interval, and other clinical outcomes, even after adjusting for risk factors such as age, tumor size, HER2 status, and PAM50 subtypes. Cosinet represents a remarkable development toward unlocking the potential of DCE analysis in the context of precision medicine.

Applying differential network analysis to longitudinal gene expression in response to perturbations

Differential Network (DN) analysis is a method that has long been used to interpret changes in gene expression data and provide biological insights. The method identifies the rewiring of gene networks in response to external perturbations. Our study applies the DN method to the analysis of RNA-sequencing (RNA-seq) time series datasets. We focus on expression changes: (i) in saliva of a human subject after pneumococcal vaccination (PPSV23) and (ii) in primary B cells treated ex vivo with a monoclonal antibody drug (Rituximab). The DN method enabled us to identify the activation of biological pathways consistent with the mechanisms of action of the PPSV23 vaccine and target pathways of Rituximab. The community detection algorithm on the DN revealed clusters of genes characterized by collective temporal behavior. All saliva and some B cell DN communities showed characteristic time signatures, outlining a chronological order in pathway activation in response to the perturbation. Moreover, we identified early and delayed responses within network modules in the saliva dataset and three temporal patterns in the B cell data.

Estimation of Gene Regulatory Networks from Cancer Transcriptomics Data

Cancer is a genetic disease in which multiple genes are perturbed. Thus, information about the regulatory relationships between genes is necessary for the identification of biomarkers and therapeutic targets. In this review, methods for inference of gene regulatory networks (GRNs) from transcriptomics data that are used in cancer research are introduced. The methods are classified into three categories according to the analysis model. The first category includes methods that use pair-wise measures between genes, including correlation coefficient and mutual information. The second category includes methods that determine the genetic regulatory relationship using multivariate measures, which consider the expression profiles of all genes concurrently. The third category includes methods using supervised and integrative approaches. The supervised approach estimates the regulatory relationship using a supervised learning method that constructs a regression or classification model for predicting whether there is a regulatory relationship between genes with input data of gene expression profiles and class labels of prior biological knowledge. The integrative method is an expansion of the supervised method and uses more data and biological knowledge for predicting the regulatory relationship. Furthermore, simulation and experimental validation of the estimated GRNs are also discussed in this review. This review identified that most GRN inference methods are not specific for cancer transcriptome data, and such methods are required for better understanding of cancer pathophysiology. In addition, more systematic methods for validation of the estimated GRNs need to be developed in the context of cancer biology.

Differential Co-Expression Analyses Allow the Identification of Critical Signalling Pathways Altered during Tumour Transformation and Progression

Biological systems respond to perturbations through the rewiring of molecular interactions, organised in gene regulatory networks (GRNs). Among these, the increasingly high availability of transcriptomic data makes gene co-expression networks the most exploited ones. Differential co-expression networks are useful tools to identify changes in response to an external perturbation, such as mutations predisposing to cancer development, and leading to changes in the activity of gene expression regulators or signalling. They can help explain the robustness of cancer cells to perturbations and identify promising candidates for targeted therapy, moreover providing higher specificity with respect to standard co-expression methods. Here, we comprehensively review the literature about the methods developed to assess differential co-expression and their applications to cancer biology. Via the comparison of normal and diseased conditions and of different tumour stages, studies based on these methods led to the definition of pathways involved in gene network reorganisation upon oncogenes' mutations and tumour progression, often converging on immune system signalling. A relevant implementation still lagging behind is the integration of different data types, which would greatly improve network interpretability. Most importantly, performance and predictivity evaluation of the large variety of mathematical models proposed would urgently require experimental validations and systematic comparisons. We believe that future work on differential gene co-expression networks, complemented with additional omics data and experimentally tested, will considerably improve our insights into the biology of tumours.

An integrative multi-omics network-based approach identifies key regulators for breast cancer

Although genome-wide association studies (GWASs) have successfully identified thousands of risk variants for human complex diseases, understanding the biological function and molecular mechanisms of the associated SNPs involved in complex diseases is challenging. Here we developed a framework named integrative multi-omics network-based approach (IMNA), aiming to identify potential key genes in regulatory networks by integrating molecular interactions across multiple biological scales, including GWAS signals, gene expression-based signatures, chromatin interactions and protein interactions from the network topology. We applied this approach to breast cancer, and prioritized key genes involved in regulatory networks. We also developed an abnormal gene expression score (AGES) signature based on the gene expression deviation of the top 20 rank-ordered genes in breast cancer. The AGES values are associated with genetic variants, tumor properties and patient survival outcomes. Among the top 20 genes, RNASEH2A was identified as a new candidate gene for breast cancer. Thus, our integrative network-based approach provides a genetic-driven framework to unveil tissue-specific interactions from multiple biological scales and reveal potential key regulatory genes for breast cancer. This approach can also be applied in other complex diseases such as ovarian cancer to unravel underlying mechanisms and help for developing therapeutic targets.

Classification of Cancer Types Using Graph Convolutional Neural Networks

BACKGROUND

Cancer has been a leading cause of death in the United States with significant health care costs. Accurate prediction of cancers at an early stage and understanding the genomic mechanisms that drive cancer development are vital to the improvement of treatment outcomes and survival rates, thus resulting in significant social and economic impacts. Attempts have been made to classify cancer types with machine learning techniques during the past two decades and deep learning approaches more recently.

RESULTS

In this paper, we established four models with graph convolutional neural network (GCNN) that use unstructured gene expressions as inputs to classify different tumor and non-tumor samples into their designated 33 cancer types or as normal. Four GCNN models based on a co-expression graph, co-expression+singleton graph, protein-protein interaction (PPI) graph, and PPI+singleton graph have been designed and implemented. They were trained and tested on combined 10,340 cancer samples and 731 normal tissue samples from The Cancer Genome Atlas (TCGA) dataset. The established GCNN models achieved excellent prediction accuracies (89.9-94.7%) among 34 classes (33 cancer types and a normal group). In silico gene-perturbation experiments were performed on four models based on co-expression graph, co-expression+singleton, PPI graph, and PPI+singleton graphs. The co-expression GCNN model was further interpreted to identify a total of 428 markers genes that drive the classification of 33 cancer types and normal. The concordance of differential expressions of these markers between the represented cancer type and others are confirmed. Successful classification of cancer types and a normal group regardless of normal tissues' origin suggested that the identified markers are cancer-specific rather than tissue-specific.

CONCLUSION

Novel GCNN models have been established to predict cancer types or normal tissue based on gene expression profiles. We demonstrated the results from the TCGA dataset that these models can produce accurate classification (above 94%), using cancer-specific markers genes. The models and the source codes are publicly available and can be readily adapted to the diagnosis of cancer and other diseases by the data-driven modeling research community.

Computational medicine: quantitative modeling of complex diseases

Biological complex systems are composed of numerous components that interact within and across different scales. The ever-increasing generation of high-throughput biomedical data has given us an opportunity to develop a quantitative model of nonlinear biological systems having implications in health and diseases. Multidimensional molecular data can be modeled using various statistical methods at different scales of biological organization, such as genome, transcriptome and proteome. I will discuss recent advances in the application of computational medicine in complex diseases such as network-based studies, genome-scale metabolic modeling, kinetic modeling and support vector machines with specific examples in the field of cancer, psychiatric disorders and type 2 diabetes. The recent advances in translating these computational models in diagnosis and identification of drug targets of complex diseases are discussed, as well as the challenges researchers and clinicians are facing in taking computational medicine from the bench to bedside.

Differential co-expression-based detection of conditional relationships in transcriptional data: comparative analysis and application to breast cancer

BACKGROUND

Elucidation of regulatory networks, including identification of regulatory mechanisms specific to a given biological context, is a key aim in systems biology. This has motivated the move from co-expression to differential co-expression analysis and numerous methods have been developed subsequently to address this task; however, evaluation of methods and interpretation of the resulting networks has been hindered by the lack of known context-specific regulatory interactions.

RESULTS

In this study, we develop a simulator based on dynamical systems modelling capable of simulating differential co-expression patterns. With the simulator and an evaluation framework, we benchmark and characterise the performance of inference methods. Defining three different levels of "true" networks for each simulation, we show that accurate inference of causation is difficult for all methods, compared to inference of associations. We show that a z-score-based method has the best general performance. Further, analysis of simulation parameters reveals five network and simulation properties that explained the performance of methods. The evaluation framework and inference methods used in this study are available in the dcanr R/Bioconductor package.

CONCLUSIONS

Our analysis of networks inferred from simulated data show that hub nodes are more likely to be differentially regulated targets than transcription factors. Based on this observation, we propose an interpretation of the inferred differential network that can reconstruct a putative causal network.

Multi-omics analysis reveals the BRCA1 mutation and mismatch repair gene signatures associated with survival, protein expression, and copy number alterations in prostate cancer

BACKGROUND

Recent genomic analysis reveals that DNA repair gene mutations can be detected in 15-30% of patients in metastatic castration resistant prostate cancer depending on the population and clinical setting when comparing to a very small fraction in those with indolent localized diseases. The discovery and characterization of function associated with DNA repair gene mutations in prostate cancer patients may increase therapeutic options and lead to improved clinical outcomes.

METHODS

To understand the role of DNA repair genes associated with other genomic alteration and signaling pathway, we applied an integrative analysis of multi-omics to The Cancer Genome Atlas (TCGA) prostate cancer dataset which contains 498 patients. We concurrently analyzed gene expression profiles, reverse phase protein lysate microarray (RPPA) data, and copy number alterations to examine the potential genomic mechanisms.

RESULTS

We identified the signature of "chromosome condensation", "BRCA1 mutation", and "mismatch repair" were associated with disease-free survival in prostate cancer. Through the concurrent analysis of gene expression profiles, reverse RPPA data, and copy number alterations, we found the three signatures are associated with cell cycle and DNA repair pathway and also most events of copy number gains.

CONCLUSIONS

This study presents a unique extension from DNA mutations to expressional functions, proteomic activities, and copy numbers of DNA repair genes in prostate cancer. Our findings revealed crucial prognostic markers and candidates for further biological and clinical investigations.

scdNet: a computational tool for single-cell differential network analysis

Background

Single-cell RNA sequencing (scRNA-Seq) is an emerging technology that has revolutionized the research of the tumor heterogeneity. However, the highly sparse data matrices generated by the technology have posed an obstacle to the analysis of differential gene regulatory networks.

Results

Addressing the challenges, this study presents, as far as we know, the first bioinformatics tool for scRNA-Seq-based differential network analysis (scdNet). The tool features a sample size adjustment of gene-gene correlation, comparison of inter-state correlations, and construction of differential networks. A simulation analysis demonstrated the power of scdNet in the analyses of sparse scRNA-Seq data matrices, with low requirement on the sample size, high computation efficiency, and tolerance of sequencing noises. Applying the tool to analyze two datasets of single circulating tumor cells (CTCs) of prostate cancer and early mouse embryos, our data demonstrated that differential gene regulation plays crucial roles in anti-androgen resistance and early embryonic development.

Conclusions

Overall, the tool is widely applicable to datasets generated by the emerging technology to bring biological insights into tumor heterogeneity and other studies. MATLAB implementation of scdNet is available at https://github.com/ChenLabGCCRI/scdNet.

GSAE: an autoencoder with embedded gene-set nodes for genomics functional characterization

Background

Bioinformatics tools have been developed to interpret gene expression data at the gene set level, and these gene set based analyses improve the biologists’ capability to discover functional relevance of their experiment design. While elucidating gene set individually, inter-gene sets association is rarely taken into consideration. Deep learning, an emerging machine learning technique in computational biology, can be used to generate an unbiased combination of gene set, and to determine the biological relevance and analysis consistency of these combining gene sets by leveraging large genomic data sets.

Results

In this study, we proposed a gene superset autoencoder (GSAE), a multi-layer autoencoder model with the incorporation of a priori defined gene sets that retain the crucial biological features in the latent layer. We introduced the concept of the gene superset, an unbiased combination of gene sets with weights trained by the autoencoder, where each node in the latent layer is a superset. Trained with genomic data from TCGA and evaluated with their accompanying clinical parameters, we showed gene supersets’ ability of discriminating tumor subtypes and their prognostic capability. We further demonstrated the biological relevance of the top component gene sets in the significant supersets.

Conclusions

Using autoencoder model and gene superset at its latent layer, we demonstrated that gene supersets retain sufficient biological information with respect to tumor subtypes and clinical prognostic significance. Superset also provides high reproducibility on survival analysis and accurate prediction for cancer subtypes.

Electronic supplementary material

The online version of this article (10.1186/s12918-018-0642-2) contains supplementary material, which is available to authorized users.

FACS-Seq analysis of Pax3-derived cells identifies non-myogenic lineages in the embryonic forelimb

Skeletal muscle in the forelimb develops during embryonic and fetal development and perinatally. While much is known regarding the molecules involved in forelimb myogenesis, little is known about the specific mechanisms and interactions. Migrating skeletal muscle precursor cells express Pax3 as they migrate into the forelimb from the dermomyotome. To compare gene expression profiles of the same cell population over time, we isolated lineage-traced Pax3⁺ cells (Pax3^EGFP) from forelimbs at different embryonic days. We performed whole transcriptome profiling via RNA-Seq of Pax3⁺ cells to construct gene networks involved in different stages of embryonic and fetal development. With this, we identified genes involved in the skeletal, muscular, vascular, nervous and immune systems. Expression of genes related to the immune, skeletal and vascular systems showed prominent increases over time, suggesting a non-skeletal myogenic context of Pax3-derived cells. Using co-expression analysis, we observed an immune-related gene subnetwork active during fetal myogenesis, further implying that Pax3-derived cells are not a strictly myogenic lineage, and are involved in patterning and three-dimensional formation of the forelimb through multiple systems.

Differential gene regulatory networks in development and disease

Gene regulatory networks, in which differential expression of regulator genes induce differential expression of their target genes, underlie diverse biological processes such as embryonic development, organ formation and disease pathogenesis. An archetypical systems biology approach to mapping these networks involves the combined application of (1) high-throughput sequencing-based transcriptome profiling (RNA-seq) of biopsies under diverse network perturbations and (2) network inference based on gene-gene expression correlation analysis. The comparative analysis of such correlation networks across cell types or states, differential correlation network analysis, can identify specific molecular signatures and functional modules that underlie the state transition or have context-specific function. Here, we review the basic concepts of network biology and correlation network inference, and the prevailing methods for differential analysis of correlation networks. We discuss applications of gene expression network analysis in the context of embryonic development, cancer, and congenital diseases.

Genome-wide identification of key modulators of gene-gene interaction networks in breast cancer

Background

With the advances in high-throughput gene profiling technologies, a large volume of gene interaction maps has been constructed. A higher-level layer of gene-gene interaction, namely modulate gene interaction, is composed of gene pairs of which interaction strengths are modulated by (i.e., dependent on) the expression level of a key modulator gene. Systematic investigations into the modulation by estrogen receptor (ER), the best-known modulator gene, have revealed the functional and prognostic significance in breast cancer. However, a genome-wide identification of key modulator genes that may further unveil the landscape of modulated gene interaction is still lacking.

Results

We proposed a systematic workflow to screen for key modulators based on genome-wide gene expression profiles. We designed four modularity parameters to measure the ability of a putative modulator to perturb gene interaction networks. Applying the method to a dataset of 286 breast tumors, we comprehensively characterized the modularity parameters and identified a total of 973 key modulator genes. The modularity of these modulators was verified in three independent breast cancer datasets. ESR1, the encoding gene of ER, appeared in the list, and abundant novel modulators were illuminated. For instance, a prognostic predictor of breast cancer, SFRP1, was found the second modulator. Functional annotation analysis of the 973 modulators revealed involvements in ER-related cellular processes as well as immune- and tumor-associated functions.

Conclusions

Here we present, as far as we know, the first comprehensive analysis of key modulator genes on a genome-wide scale. The validity of filtering parameters as well as the conservativity of modulators among cohorts were corroborated. Our data bring new insights into the modulated layer of gene-gene interaction and provide candidates for further biological investigations.

Ovarian Cancer Differential Interactome and Network Entropy Analysis Reveal New Candidate Biomarkers

Ovarian cancer is one of the most common cancers and has a high mortality rate due to insidious symptoms and lack of robust diagnostics. A hitherto understudied concept in cancer pathogenesis may offer new avenues for innovation in ovarian cancer biomarker development. Cancer cells are characterized by an increase in network entropy, and several studies have exploited this concept to identify disease-associated gene and protein modules. We report in this study the changes in protein-protein interactions (PPIs) in ovarian cancer within a differential network (interactome) analysis framework utilizing the entropy concept and gene expression data. A compendium of six transcriptome datasets that included 140 samples from laser microdissected epithelial cells of ovarian cancer patients and 51 samples from healthy population was obtained from Gene Expression Omnibus, and the high confidence human protein interactome (31,465 interactions among 10,681 proteins) was used. The uncertainties of the up- or downregulation of PPIs in ovarian cancer were estimated through an entropy formulation utilizing combined expression levels of genes, and the interacting protein pairs with minimum uncertainty were identified. We identified 105 proteins with differential PPI patterns scattered in 11 modules, each indicating significantly affected biological pathways in ovarian cancer such as DNA repair, cell proliferation-related mechanisms, nucleoplasmic translocation of estrogen receptor, extracellular matrix degradation, and inflammation response. In conclusion, we suggest several PPIs as biomarker candidates for ovarian cancer and discuss their future biological implications as potential molecular targets for pharmaceutical development as well. In addition, network entropy analysis is a concept that deserves greater research attention for diagnostic innovation in oncology and tumor pathogenesis.

Higher HOPX expression is associated with distinct clinical and biological features and predicts poor prognosis in de novo acute myeloid leukemia

Homeodomain-only protein homeobox (HOPX) is the smallest homeodomain protein. It was regarded as a stem cell marker in several non-hematopoietic systems. While the prototypic homeobox genes such as the HOX family have been well characterized in acute myeloid leukemia (AML), the clinical and biological implications of HOPX in the disease remain unknown. Thus we analyzed HOPX and global gene expression patterns in 347 newly diagnosed de novo AML patients in our institute. We found that higher HOPX expression was closely associated with older age, higher platelet counts, lower white blood cell counts, lower lactate dehydrogenase levels, and mutations in RUNX1, IDH2, ASXL1, and DNMT3A, but negatively associated with acute promyelocytic leukemia, favorable karyotypes, CEBPA double mutations and NPM1 mutation. Patients with higher HOPX expression had a lower complete remission rate and shorter survival. The finding was validated in two independent cohorts. Multivariate analysis revealed that higher HOPX expression was an independent unfavorable prognostic factor irrespective of other known prognostic parameters and gene signatures derived from multiple cohorts. Gene set enrichment analysis showed higher HOPX expression was associated with both hematopoietic and leukemia stem cell signatures. While HOPX and HOX family genes showed concordant expression patterns in normal hematopoietic stem/progenitor cells, their expression patterns and associated clinical and biological features were distinctive in AML settings, demonstrating HOPX to be a unique homeobox gene. Therefore, HOPX is a distinctive homeobox gene with characteristic clinical and biological implications and its expression is a powerful predictor of prognosis in AML patients.

Differential correlation analysis of glioblastoma reveals immune ceRNA interactions predictive of patient survival

Background

Recent studies illuminated a novel role of microRNA (miRNA) in the competing endogenous RNA (ceRNA) interaction: two genes (ceRNAs) can achieve coexpression by competing for a pool of common targeting miRNAs. Individual biological investigations implied ceRNA interaction performs crucial oncogenic/tumor suppressive functions in glioblastoma multiforme (GBM). Yet, a systematic analysis has not been conducted to explore the functional landscape and prognostic significance of ceRNA interaction.

Results

Incorporating the knowledge that ceRNA interaction is highly condition-specific and modulated by the expressional abundance of miRNAs, we devised a ceRNA inference by differential correlation analysis to identify the miRNA-modulated ceRNA pairs. Analyzing sample-paired miRNA and gene expression profiles of GBM, our data showed that this alternative layer of gene interaction is essential in global information flow. Functional annotation analysis revealed its involvement in activated processes in brain, such as synaptic transmission, as well as critical tumor-associated functions. Notably, a systematic survival analysis suggested the strength of ceRNA-ceRNA interactions, rather than expressional abundance of individual ceRNAs, among three immune response genes (CCL22, IL2RB, and IRF4) is predictive of patient survival. The prognostic value was validated in two independent cohorts.

Conclusions

This work addresses the lack of a comprehensive exploration into the functional and prognostic relevance of ceRNA interaction in GBM. The proposed efficient and reliable method revealed its significance in GBM-related functions and prognosis. The highlighted roles of ceRNA interaction provide a basis for further biological and clinical investigations.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-017-1557-4) contains supplementary material, which is available to authorized users.