Gene Subnetworks Research Articles

A component overlapping attribute clustering (COAC) algorithm for single-cell RNA sequencing data analysis and potential pathobiological implications.

Recent advances in next-generation sequencing and computational technologies have enabled routine analysis of large-scale single-cell ribonucleic acid sequencing (scRNA-seq) data. However, scRNA-seq technologies have suffered from several technical challenges, including low mean expression levels in most genes and higher frequencies of missing data than bulk population sequencing technologies. Identifying functional gene sets and their regulatory networks that link specific cell types to human diseases and therapeutics from scRNA-seq profiles are daunting tasks. In this study, we developed a Component Overlapping Attribute Clustering (COAC) algorithm to perform the localized (cell subpopulation) gene co-expression network analysis from large-scale scRNA-seq profiles. Gene subnetworks that represent specific gene co-expression patterns are inferred from the components of a decomposed matrix of scRNA-seq profiles. We showed that single-cell gene subnetworks identified by COAC from multiple time points within cell phases can be used for cell type identification with high accuracy (83%). In addition, COAC-inferred subnetworks from melanoma patients’ scRNA-seq profiles are highly correlated with survival rate from The Cancer Genome Atlas (TCGA). Moreover, the localized gene subnetworks identified by COAC from individual patients’ scRNA-seq data can be used as pharmacogenomics biomarkers to predict drug responses (The area under the receiver operating characteristic curves ranges from 0.728 to 0.783) in cancer cell lines from the Genomics of Drug Sensitivity in Cancer (GDSC) database. In summary, COAC offers a powerful tool to identify potential network-based diagnostic and pharmacogenomics biomarkers from large-scale scRNA-seq profiles. COAC is freely available at https://github.com/ChengF-Lab/COAC.

Classifying mild traumatic brain injuries with functional network analysis

BackgroundTraumatic brain injury (TBI) represents a critical health problem of which timely diagnosis and treatment remain challenging. TBI is a result of an external force damaging brain tissue, accompanied by delayed pathogenic events which aggravate the injury. Molecular responses to different mild TBI subtypes have not been well characterized. TBI subtype classification is an important step towards the development and application of novel treatments. The computational systems biology approach is proved to be a promising tool in biomarker discovery for central nervous system injury.ResultsIn this study, we have performed a network-based analysis on gene expression profiles to identify functional gene subnetworks. The gene expression profiles were obtained from two experimental models of injury in rats: the controlled cortical impact and the fluid percussion injury. Our method integrates protein interaction information with gene expression profiles to identify subnetworks of genes as biomarkers. We have demonstrated that the selected gene subnetworks are more accurate to classify the heterogeneous responses to different injury models, compared to conventional analysis using individual marker genes selected without network information.ConclusionsThe systems approach can lead to a better understanding of the underlying complexities of the molecular responses after TBI and the identified subnetworks will have important prognostic functions for patients who sustain mild TBIs.

GWAS and network analysis of co-occurring nicotine and alcohol dependence identifies significantly associated alleles and network.

Alcohol dependence (AD) and nicotine dependence (ND) co-occur frequently (AD+ND). We integrated SNP-based, gene-based, and protein-protein interaction network analyses to identify shared risk genes or gene subnetworks for AD+ND in African Americans (AAs, N = 2,094) and European Americans (EAs, N = 1,207). The DSM-IV criterion counts for AD and ND were modeled as two dependent variables in a multivariate linear mixed model, and analyzed separately for the two populations. The most significant SNP was rs6579845 in EAs (p < 1.29 × 10-8 ) in GM2A, which encodes GM2 ganglioside activator, and is a cis-expression quantitative locus that affects GM2A expression in blood and brain tissues. However, this SNP was not replicated in our another small sample (N = 678). We identified a subnetwork of 24 genes that contributed to the AD+ND criterion counts. In the gene-set analysis for the subnetwork in an independent sample, the Study of Addiction: Genetics and Environment project (predominately EAs), these 24 genes as a set differed in AD+ND versus control subjects in EAs (p = .041). Functional enrichment analysis for this subnetwork revealed that the gene enrichment involved primarily nerve growth factor pathways, and cocaine and amphetamine addiction. In conclusion, we identified a genome-wide significant variant at GM2A and a gene subnetwork underlying the genetic trait of shared AD+ND. These results increase our understanding of the shared (pleiotropic) genetic risk that underlies AD+ND.

Identification of highly connected and differentially expressed gene subnetworks in metastasizing endometrial cancer.

We have identified nine highly connected and differentially expressed gene subnetworks between aggressive primary tumors and metastatic lesions in endometrial carcinomas. We implemented a novel pipeline combining gene set and network approaches, which here allows integration of protein-protein interactions and gene expression data. The resulting subnetworks are significantly associated with disease progression across tumor stages from complex atypical hyperplasia, primary tumors to metastatic lesions. The nine subnetworks include genes related to metastasizing features such as epithelial-mesenchymal transition (EMT), hypoxia and cell proliferation. TCF4 and TWIST2 were found as central genes in the subnetwork related to EMT. Two of the identified subnetworks display statistically significant association to patient survival, which were further supported by an independent validation in the data from The Cancer Genome Atlas data collection. The first subnetwork contains genes related to cell proliferation and cell cycle, while the second contains genes involved in hypoxia such as HIF1A and EGLN3. Our findings provide a promising context to elucidate the biological mechanisms of metastasis, suggest potential prognostic markers and further identify therapeutic targets. The pipeline R source code is freely available, including permutation tests to assess statistical significance of the identified subnetworks.

A Comparison of the TempO-Seq S1500+ Platform to RNA-Seq and Microarray Using Rat Liver Mode of Action Samples.

The TempO-SeqTM platform allows for targeted transcriptomic analysis and is currently used by many groups to perform high-throughput gene expression analysis. Herein we performed a comparison of gene expression characteristics measured using 45 purified RNA samples from the livers of rats exposed to chemicals that fall into one of five modes of action (MOAs). These samples have been previously evaluated using AffymetrixTM rat genome 230 2.0 microarrays and Illumina® whole transcriptome RNA-Seq. Comparison of these data with TempO-Seq analysis using the rat S1500+ beta gene set identified clear differences in the platforms related to signal to noise, root mean squared error, and/or sources of variability. Microarray and TempO-Seq captured the most variability in terms of MOA and chemical treatment whereas RNA-Seq had higher noise and larger differences between samples within a MOA. However, analysis of the data by hierarchical clustering, gene subnetwork connectivity and biological process representation of MOA-varying genes revealed that the samples clearly grouped by treatment as opposed to gene expression platform. Overall these findings demonstrate that the results from the TempO-Seq platform are consistent with findings on other more established approaches for measuring the genome-wide transcriptome.

Integrated systems analysis reveals conserved gene networks underlying response to spinal cord injury.

Spinal cord injury (SCI) is a devastating neurological condition for which there are currently no effective treatment options to restore function. A major obstacle to the development of new therapies is our fragmentary understanding of the coordinated pathophysiological processes triggered by damage to the human spinal cord. Here, we describe a systems biology approach to integrate decades of small-scale experiments with unbiased, genome-wide gene expression from the human spinal cord, revealing a gene regulatory network signature of the pathophysiological response to SCI. Our integrative analyses converge on an evolutionarily conserved gene subnetwork enriched for genes associated with the response to SCI by small-scale experiments, and whose expression is upregulated in a severity-dependent manner following injury and downregulated in functional recovery. We validate the severity-dependent upregulation of this subnetwork in rodents in primary transcriptomic and proteomic studies. Our analysis provides systems-level view of the coordinated molecular processes activated in response to SCI.

A novel approach to identify highly connected and differentially expressed gene subnetworks in metastasizing endometrial cancer

A novel approach to identify highly connected and differentially expressed gene subnetworks in metastasizing endometrial cancer

Abstract 1141: A novel regulatory gene subnetwork of glioma stem-like cells

Abstract Glioblastoma (GBM) is the most common and deadliest malignant brain tumor in adults. GBM is enriched in glioma stem-like cells (GSCs) that contribute to tumor initiation and treatment resistance and thus a natural focus for therapeutic development. However, targeting GSCs has been a challenge because of a lack of master regulators specific to GSCs and not to normal brain cells. Therefore, systematic identification of the GSC-specific common regulatory program will improve our understanding of GSCs and provide an opportunity to develop GSC-specific therapeutics that are efficacious and safe. To accomplish these objectives, we have recently developed two innovative tandem computational platforms, GeneRep and nSCORE, which were subsequently experimentally validated. The tandem platforms complement each other and allow us to 1) interrogate large-scale gene expression profiles of GSCs, GBM-differentiated cells, normal neuronal precursor cells (NPCs) and astrocytes (NAs) to extract candidate GSC-specific gene regulatory networks; and 2) functionally rank GSC-specific core master regulators, respectively. A cluster of 8 interrelated master subnetworks emerged, which were functionally grouped into the stemness and cancer pathways, some of which belonged to the Sox, Olig and Myc families. By enforcing the expression of this cluster in NAs, we were able to reprogram these normal cells into GSCs as measured by the neurosphere formation and in vivo tumorigenesis assays, confirming the critical requirement of this cluster for the GSC identity. Furthermore, this network cluster was present and functioned cooperatively to maintain GSCs in several patient-derived GSC lines examined to date. Survival and neurosphere formation rates of GSCs were profoundly suppressed when at least two nodes in this cluster were depleted by siRNA compared to depletion of the individual nodes. Our data establish a novel gene regulatory cluster controlling GSCs and provide a compelling rationale for its therapeutic targeting to reduce tumor recurrence. Citation Format: Changwang Deng, Son B. Le, David D. Tran. A novel regulatory gene subnetwork of glioma stem-like cells [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 1141.

Developmental Chromatin Restriction of Pro-Growth Gene Networks Acts as an Epigenetic Barrier to Axon Regeneration in Cortical Neurons.

Axon regeneration in the central nervous system is prevented in part by a developmental decline in the intrinsic regenerative ability of maturing neurons. This loss of axon growth ability likely reflects widespread changes in gene expression, but the mechanisms that drive this shift remain unclear. Chromatin accessibility has emerged as a key regulatory mechanism in other cellular contexts, raising the possibility that chromatin structure may contribute to the age-dependent loss of regenerative potential. Here we establish an integrated bioinformatic pipeline that combines analysis of developmentally dynamic gene networks with transcription factor regulation and genome-wide maps of chromatin accessibility. When applied to the developing cortex, this pipeline detected overall closure of chromatin in sub-networks of genes associated with axon growth. We next analyzed mature CNS neurons that were supplied with various pro-regenerative transcription factors. Unlike prior results with SOX11 and KLF7, here we found that neither JUN nor an activated form of STAT3 promoted substantial corticospinal tract regeneration. Correspondingly, chromatin accessibility in JUN or STAT3 target genes was substantially lower than in predicted targets of SOX11 and KLF7. Finally, we used the pipeline to predict pioneer factors that could potentially relieve chromatin constraints at growth-associated loci. Overall this integrated analysis substantiates the hypothesis that dynamic chromatin accessibility contributes to the developmental decline in axon growth ability and influences the efficacy of pro-regenerative interventions in the adult, while also pointing toward selected pioneer factors as high-priority candidates for future combinatorial experiments. © 2018 Wiley Periodicals, Inc. Develop Neurobiol 00: 000-000, 2018.

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 (Pax3EGFP) 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.

Identification of key candidate genes and pathways in hepatocellular carcinoma by integrated bioinformatical analysis.

Hepatocellular carcinoma (HCC) is one of the most common malignant neoplasms worldwide, however the underlying mechanisms and gene signatures of HCC are unknown. In the present study the profile datasets of four cohorts were integrated to elucidate the pathways and candidate genes of HCC. The expression profiles GSE25097, GSE45267, GSE57957 and GSE62232 were downloaded from the Gene Expression Omnibus database, including 436 HCC and 94 normal liver tissues. A total of 185 differentially expressed genes (DEGs) were identified in HCC, including 92 upregulated genes and 92 downregulated genes. Gene ontology (GO) was performed, which revealed that the upregulated DEGs were primarily enriched in cell division, mitotic nuclear division, mitotic cytokinesis and G1/S transition of the mitotic cell cycle. Pathway enrichment was analyzed based on the Kyoto Encyclopedia of Genes and Genomes database to assess the functional relevance of DEGs. The most significant module was selected from protein-protein interactions and 15 important hub genes were identified. The sub-networks of hub genes were involved in cell division, p53 signaling, and T lymphotropic virus type I infection signaling pathways. In conclusion, the present study revealed that the identified DEG candidate genes may promote the understanding of the cause and molecular mechanisms underlying the development of HCC and that these candidates and signal pathways may be potential targets of clinical therapy for HCC.

SPSNet: subpopulation-sensitive network-based analysis of heterogeneous gene expression data

BackgroundTranscriptomic datasets often contain undeclared heterogeneity arising from biological variation such as diversity of disease subtypes, treatment subgroups, time-series gene expression, nested experimental conditions, as well as technical variation due to batch effects, platform differences in integrated meta-analyses, etc. However, current analysis approaches are primarily designed to handle comparisons between experimental conditions represented by homogeneous samples, thus precluding the discovery of underlying subphenotypes. Unsupervised methods for subtype identification are typically based on individual gene level analysis, which often result in irreproducible gene signatures for potential subtypes. Emerging methods to study heterogeneity have been largely developed in the context of single-cell datasets containing hundreds to thousands of samples, limiting their use to select contexts.ResultsWe present a novel analysis method, SPSNet, which identifies subtype-specific gene expression signatures based on the activity of subnetworks in biological pathways. SPSNet identifies the gene subnetworks capturing the diversity of underlying biological mechanisms, indicating potential sample subphenotypes. In the presence of extrinsic or non-biological heterogeneity (e.g. batch effects), SPSNet identifies subnetworks that are particularly affected by such variation, thus helping eliminate factors irrelevant to the biology of the phenotypes under study.ConclusionUsing multiple publicly available datasets, we illustrate that SPSNet is able to consistently uncover patterns within gene expression data that correspond to meaningful heterogeneity of various origins. We also demonstrate the performance of SPSNet as a sensitive and reliable tool for understanding the structure and nature of such heterogeneity.

Identifying noncoding risk variants using disease-relevant gene regulatory networks

Identifying noncoding risk variants remains a challenging task. Because noncoding variants exert their effects in the context of a gene regulatory network (GRN), we hypothesize that explicit use of disease-relevant GRNs can significantly improve the inference accuracy of noncoding risk variants. We describe Annotation of Regulatory Variants using Integrated Networks (ARVIN), a general computational framework for predicting causal noncoding variants. It employs a set of novel regulatory network-based features, combined with sequence-based features to infer noncoding risk variants. Using known causal variants in gene promoters and enhancers in a number of diseases, we show ARVIN outperforms state-of-the-art methods that use sequence-based features alone. Additional experimental validation using reporter assay further demonstrates the accuracy of ARVIN. Application of ARVIN to seven autoimmune diseases provides a holistic view of the gene subnetwork perturbed by the combinatorial action of the entire set of risk noncoding mutations.

Integrated Bayesian analysis of rare exonic variants to identify risk genes for schizophrenia and neurodevelopmental disorders

BackgroundIntegrating rare variation from trio family and case–control studies has successfully implicated specific genes contributing to risk of neurodevelopmental disorders (NDDs) including autism spectrum disorders (ASD), intellectual disability (ID), developmental disorders (DDs), and epilepsy (EPI). For schizophrenia (SCZ), however, while sets of genes have been implicated through the study of rare variation, only two risk genes have been identified.MethodsWe used hierarchical Bayesian modeling of rare-variant genetic architecture to estimate mean effect sizes and risk-gene proportions, analyzing the largest available collection of whole exome sequence data for SCZ (1,077 trios, 6,699 cases, and 13,028 controls), and data for four NDDs (ASD, ID, DD, and EPI; total 10,792 trios, and 4,058 cases and controls).ResultsFor SCZ, we estimate there are 1,551 risk genes. There are more risk genes and they have weaker effects than for NDDs. We provide power analyses to predict the number of risk-gene discoveries as more data become available. We confirm and augment prior risk gene and gene set enrichment results for SCZ and NDDs. In particular, we detected 98 new DD risk genes at FDR < 0.05. Correlations of risk-gene posterior probabilities are high across four NDDs (ρ>0.55), but low between SCZ and the NDDs (ρ<0.3). An in-depth analysis of 288 NDD genes shows there is highly significant protein–protein interaction (PPI) network connectivity, and functionally distinct PPI subnetworks based on pathway enrichment, single-cell RNA-seq cell types, and multi-region developmental brain RNA-seq.ConclusionsWe have extended a pipeline used in ASD studies and applied it to infer rare genetic parameters for SCZ and four NDDs (https://github.com/hoangtn/extTADA). We find many new DD risk genes, supported by gene set enrichment and PPI network connectivity analyses. We find greater similarity among NDDs than between NDDs and SCZ. NDD gene subnetworks are implicated in postnatally expressed presynaptic and postsynaptic genes, and for transcriptional and post-transcriptional gene regulation in prenatal neural progenitor and stem cells.

Identification of a histone family gene signature for predicting the prognosis of cervical cancer patients

Heterogeneity in terms of tumor characteristics, prognosis, and survival among cancer patients is an unsolved issue. Here, we systematically analyzed the aberrant expression patterns of cervical cancer using RNA-Seq data from The Cancer Genome Atlas (TCGA). We incorporated gene profiling, molecular signatures, functional and pathway information with gene set enrichment and protein-protein interaction (PPI) network analysis, to identify sub-networks of genes. Those identified genes relating to DNA replication and DNA repair-mediated signaling pathways associated with systemic lupus erythematosus (SLE). Next, we combined cross-validated prognostic scores to build an integrated prognostic model for survival prediction. The combined approach revealed that the DNA repair-mediated including the functional interaction module of 18 histone genes (Histone cluster 1 H2A, B and H4), were significantly correlated with the survival rate. Furthermore, five of these histone genes were highly expressed in three cervical cancer cohorts from the Oncomine database. Comparison of high and low histone variant-expressing human cervical cancer cell lines revealed different responses to DNA damage, suggesting protective functions of histone genes against DNA damage. Collectively, we provide evidence that two SLE-associated gene sets (HIST1H2BD and HIST1H2BJ; and HIST1H2BD, HIST1H2BJ, HIST1H2BH, HIST1H2AM and HIST1H4K) can be used as prognostic factors for survival prediction among cervical cancer patients.

Informed walks: whispering hints to gene hunters inside networks\u2019 jungle

BackgroundSystemic approaches offer a different point of view on the analysis of several types of molecular associations as well as on the identification of specific gene communities in several cancer types. However, due to lack of sufficient data needed to construct networks based on experimental evidence, statistical gene co-expression networks are widely used instead. Many efforts have been made to exploit the information hidden in these networks. However, these approaches still need to capitalize comprehensively the prior knowledge encrypted into molecular pathway associations and improve their efficiency regarding the discovery of both exclusive subnetworks as candidate biomarkers and conserved subnetworks that may uncover common origins of several cancer types.MethodsIn this study we present the development of the Informed Walks model based on random walks that incorporate information from molecular pathways to mine candidate genes and gene-gene links. The proposed model has been applied to TCGA (The Cancer Genome Atlas) datasets from seven different cancer types, exploring the reconstructed co-expression networks of the whole set of genes and driving to highlighted sub-networks for each cancer type. In the sequel, we elucidated the impact of each subnetwork on the indication of underlying exclusive and common molecular mechanisms as well as on the short-listing of drugs that have the potential to suppress the corresponding cancer type through a drug-repurposing pipeline.ConclusionsWe have developed a method of gene subnetwork highlighting based on prior knowledge, capable to give fruitful insights regarding the underlying molecular mechanisms and valuable input to drug-repurposing pipelines for a variety of cancer types.

Novel Gene and Network Associations Found for Acute Lymphoblastic Leukemia Using Case-Control and Family-Based Studies in Multiethnic Populations.

Background: Acute lymphoblastic leukemia (ALL) is the most common childhood cancer, suggesting that germline variants influence ALL risk. Although multiple genome-wide association (GWA) studies have identified variants predisposing children to ALL, it remains unclear whether genetic heterogeneity affects ALL susceptibility and how interactions within and among genes containing ALL-associated variants influence ALL risk.Methods: Here, we jointly analyzed two published datasets of case-control GWA summary statistics along with germline data from ALL case-parent trios. We used the gene-level association method PEGASUS to identify genes with multiple variants associated with ALL. We then used PEGASUS gene scores as input to the network analysis algorithm HotNet2 to characterize the genomic architecture of ALL.Results: Using PEGASUS, we confirmed associations previously observed at genes such as ARID5B, IKZF1, CDKN2A/2B, and PIP4K2A, and we identified novel candidate gene associations. Using HotNet2, we uncovered significant gene subnetworks that may underlie inherited ALL risk: a subnetwork involved in B-cell differentiation containing the ALL-associated gene CEBPE, and a subnetwork of homeobox genes, including MEIS1Conclusions: Gene and network analysis uncovered loci associated with ALL that are missed by GWA studies, such as MEIS1 Furthermore, ALL-associated loci do not appear to interact directly with each other to influence ALL risk, and instead appear to influence leukemogenesis through multiple, complex pathways.Impact: We present a new pipeline for post hoc analysis of association studies that yields new insight into the etiology of ALL and can be applied in future studies to shed light on the genomic underpinnings of cancer. Cancer Epidemiol Biomarkers Prev; 26(10); 1531-9. ©2017 AACR.

Drug enrichment and discovery from schizophrenia genome-wide association results: an analysis and visualisation approach

Using successful genome-wide association results in psychiatry for drug repurposing is an ongoing challenge. Databases collecting drug targets and gene annotations are growing and can be harnessed to shed a new light on psychiatric disorders. We used genome-wide association study (GWAS) summary statistics from the Psychiatric Genetics Consortium (PGC) Schizophrenia working group to build a drug repositioning model for schizophrenia. As sample size increases, schizophrenia GWAS results show increasing enrichment for known antipsychotic drugs, selective calcium channel blockers, and antiepileptics. Each of these therapeutical classes targets different gene subnetworks. We identify 123 Bonferroni-significant druggable genes outside the MHC, and 128 FDR-significant biological pathways related to neurons, synapses, genic intolerance, membrane transport, epilepsy, and mental disorders. These results suggest that, in schizophrenia, current well-powered GWAS results can reliably detect known schizophrenia drugs and thus may hold considerable potential for the identification of new therapeutic leads. Moreover, antiepileptics and calcium channel blockers may provide repurposing opportunities. This study also reveals significant pathways in schizophrenia that were not identified previously, and provides a workflow for pathway analysis and drug repurposing using GWAS results.

Network-Based Coverage of Mutational Profiles Reveals Cancer Genes.

SUMMARYA central goal in cancer genomics is to identify the somatic alterations that underpin tumor initiation and progression. While commonly mutated cancer genes are readily identifiable, those that are rarely mutated across samples are difficult to distinguish from the large numbers of other infrequently mutated genes. We introduce a method, nCOP, that considers per-individual mutational profiles within the context of protein-protein interaction networks in order to identify small connected subnetworks of genes that, while not individually frequently mutated, comprise pathways that are altered across (i.e., “cover”) a large fraction of individuals. By analyzing 6,038 samples across 24 different cancer types, we demonstrate that nCOP is highly effective in identifying cancer genes, including those with low mutation frequencies. Overall, our work demonstrates that combining per-individual mutational information with interaction networks is a powerful approach for tackling the mutational heterogeneity observed across cancers.

Detecting gene subnetworks under selection in biological pathways.

Advances in high throughput sequencing technologies have created a gap between data production and functional data analysis. Indeed, phenotypes result from interactions between numerous genes, but traditional methods treat loci independently, missing important knowledge brought by network-level emerging properties. Therefore, detecting selection acting on multiple genes affecting the evolution of complex traits remains challenging. In this context, gene network analysis provides a powerful framework to study the evolution of adaptive traits and facilitates the interpretation of genome-wide data. We developed a method to analyse gene networks that is suitable to evidence polygenic selection. The general idea is to search biological pathways for subnetworks of genes that directly interact with each other and that present unusual evolutionary features. Subnetwork search is a typical combinatorial optimization problem that we solve using a simulated annealing approach. We have applied our methodology to find signals of adaptation to high-altitude in human populations. We show that this adaptation has a clear polygenic basis and is influenced by many genetic components. Our approach, implemented in the R package signet, improves on gene-level classical tests for selection by identifying both new candidate genes and new biological processes involved in adaptation to altitude.