Presentation Title
Network-based Functional Prediction Augments Genetic Association to Predict Candidate Genes for Inflammatory Bowel Disorder in Mice
Abstract
Inflammatory bowel disorder (IBD) is a complex, heterogenous disease for which hundreds of candidate genes have been identified through genome wide association studies (GWAS). Likewise, mouse studies of IBD have identified multiple quantitative trait loci (QTL) controlling IBD susceptibility. However, integrating mouse and human genetic diversity to study molecular mechanisms of IBD remains a challenge, as risk alleles do not always align, and QTL mapping resolution is relatively low. Nevertheless, we hypothesize that human risk factors and their mouse orthologs are functionally related genes that act in the same biological processes. Therefore, to predict candidate genes influencing IBD susceptibility, we propose a novel network-based machine learning approach to positional and functional information from mouse models of IBD with human risk alleles from GWAS. Using support vector machine (SVM) classifiers with gene expression signatures in mouse immune cells, we prioritized functionally related genes associated with human IBD susceptibility. The wild-derived PWD/PhJ (PWD) and standard C57BL/6J (B6) strains show marked differences in susceptibility to two models of IBD induction. Using consomic mice carrying PWD chromosomes on the B6 background, we identified PWD chromosomes (Chrs) 1, 2, and 12 as novel loci that profoundly enhance IBD susceptibility. These loci augment previous reports of the Ccc1 locus on Chr 12 that had been identified from the Collaborative Cross strain CC011/Unc for its high penetrance of spontaneous IBD. To identify translational candidate genes in these risk loci, we trained SVM classifiers to identify subnetworks enriched with IBD GWAS genes in two tissue-specific functional networks in mice: the intestinal and hemolymphoid networks. These classifiers rank all genes in the genome by functional relatedness to IBD GWAS genes. We scored all genes on mouse Chrs 1 and 2 and the Ccc1 locus to identify network based functional candidate genes. We further enriched the functional predictions by integrating transcriptional data comparing gene expression across five different types of B6 and PWD immune cells. Finally, we integrated functional and expression-based rankings to produce a final score. A total of 47 unique genes were prioritized as possible candidates to be pursued in follow-up studies, including Pip4k2a, Lcn10, Lgmn, and Gpr65, all of which have strong functional evidence for association to IBD. Our results demonstrate the predictive potential of network-based machine learning for candidate gene ranking across species.
Primary Faculty Mentor Name
J. Matt Mahoney
Faculty/Staff Collaborators
Karolyn G. Lahue, Alisha A. Linton, Brigitte Lavoie, Qian Fang, Mahalia M. McGill, Jessica W. Crothers, Cory Teuscher, Gary M. Mawe, Anna L. Tyler, Dimitry N. Krementsov, J. Matthew Mahoney
Status
Graduate
Student College
Graduate College
Second Student College
Larner College of Medicine
Program/Major
Neuroscience
Primary Research Category
Biological Sciences
Network-based Functional Prediction Augments Genetic Association to Predict Candidate Genes for Inflammatory Bowel Disorder in Mice
Inflammatory bowel disorder (IBD) is a complex, heterogenous disease for which hundreds of candidate genes have been identified through genome wide association studies (GWAS). Likewise, mouse studies of IBD have identified multiple quantitative trait loci (QTL) controlling IBD susceptibility. However, integrating mouse and human genetic diversity to study molecular mechanisms of IBD remains a challenge, as risk alleles do not always align, and QTL mapping resolution is relatively low. Nevertheless, we hypothesize that human risk factors and their mouse orthologs are functionally related genes that act in the same biological processes. Therefore, to predict candidate genes influencing IBD susceptibility, we propose a novel network-based machine learning approach to positional and functional information from mouse models of IBD with human risk alleles from GWAS. Using support vector machine (SVM) classifiers with gene expression signatures in mouse immune cells, we prioritized functionally related genes associated with human IBD susceptibility. The wild-derived PWD/PhJ (PWD) and standard C57BL/6J (B6) strains show marked differences in susceptibility to two models of IBD induction. Using consomic mice carrying PWD chromosomes on the B6 background, we identified PWD chromosomes (Chrs) 1, 2, and 12 as novel loci that profoundly enhance IBD susceptibility. These loci augment previous reports of the Ccc1 locus on Chr 12 that had been identified from the Collaborative Cross strain CC011/Unc for its high penetrance of spontaneous IBD. To identify translational candidate genes in these risk loci, we trained SVM classifiers to identify subnetworks enriched with IBD GWAS genes in two tissue-specific functional networks in mice: the intestinal and hemolymphoid networks. These classifiers rank all genes in the genome by functional relatedness to IBD GWAS genes. We scored all genes on mouse Chrs 1 and 2 and the Ccc1 locus to identify network based functional candidate genes. We further enriched the functional predictions by integrating transcriptional data comparing gene expression across five different types of B6 and PWD immune cells. Finally, we integrated functional and expression-based rankings to produce a final score. A total of 47 unique genes were prioritized as possible candidates to be pursued in follow-up studies, including Pip4k2a, Lcn10, Lgmn, and Gpr65, all of which have strong functional evidence for association to IBD. Our results demonstrate the predictive potential of network-based machine learning for candidate gene ranking across species.