PERCH: A Unified Framework for Disease Gene Prioritization

Abstract

To interpret genetic variants discovered from next-generation sequencing, integration of heterogeneous information is vital for success. This article describes a framework named PERCH (Polymorphism Evaluation, Ranking, and Classification for a Heritable trait), available at http://BJFengLab.org/. It can prioritize disease genes by quantitatively unifying a new deleteriousness measure called BayesDel, an improved assessment of the biological relevance of genes to the disease, a modified linkage analysis, a novel rare-variant association test, and a converted variant call quality score. It supports data that contain various combinations of extended pedigrees, trios, and case-controls, and allows for a reduced penetrance, an elevated phenocopy rate, liability classes, and covariates. BayesDel is more accurate than PolyPhen2, SIFT, FATHMM, LRT, Mutation Taster, Mutation Assessor, PhyloP, GERP++, SiPhy, CADD, MetaLR, and MetaSVM. The overall approach is faster and more powerful than the existing quantitative method pVAAST, as shown by the simulations of challenging situations in finding the missing heritability of a complex disease. This framework can also classify variants of unknown significance (variants of uncertain significance) by quantitatively integrating allele frequencies, deleteriousness, association, and co-segregation. PERCH is a versatile tool for gene prioritization in gene discovery research and variant classification in clinical genetic testing.

Keywords: co-segregation analysis; de novo mutation; functional consequence; gene association network; gene prioritization; genetic testing; rare-variant burden test; variant interpretation; variants of unknown significance; whole-exome/whole-genome/gene-panel sequencing.

Figure 1

Flowchart of gene prioritization procedures by PERCH.

Figure 2

Receiver operating characteristic (ROC) curve for the prediction of pathogenic variants in test dataset 1. Numbers in parentheses are areas under the curves (AUC).

Figure 3

Receiver operating characteristic (ROC) curve for the prediction of pathogenic variants in test dataset 2. Numbers in parentheses are areas under the curves (AUC).

Figure 4

Performance of PERCH and pVAAST in gene prioritization for complex diseases. H: BayesHLR, S: BayesSeg, G: BayesGBA, d: BayesDel; H[d]+S[d]: H+S weighted by BayesDel. MAF: minor allele frequency cutoff for low-frequency variants; Causal: the percentage of low-frequency variants that are causal; RR: relative risk of causal variants; Power: the probability of finding the causal genes among the top 20, 50, 100 and 200.