A comprehensive framework for prioritizing variants in exome sequencing studies of Mendelian diseases

Abstract

Exome sequencing strategy is promising for finding novel mutations of human monogenic disorders. However, pinpointing the casual mutation in a small number of samples is still a big challenge. Here, we propose a three-level filtration and prioritization framework to identify the casual mutation(s) in exome sequencing studies. This efficient and comprehensive framework successfully narrowed down whole exome variants to very small numbers of candidate variants in the proof-of-concept examples. The proposed framework, implemented in a user-friendly software package, named KGGSeq (http://statgenpro.psychiatry.hku.hk/kggseq), will play a very useful role in exome sequencing-based discovery of human Mendelian disease genes.

Figure 1.

The three-level filtration and prioritization framework implemented in KGGSeq.

Figure 2.

Protein–protein interaction network of MYH3 with four candidate genes. The five involved genes are in dashed circle. Each filled node denotes a gene; edges between notes indicate PPIs between protein products of the corresponding genes. Different edge colors represent the types of evidence for the association. This figure was produced by STRING (V9.0).

Figure 3.

Pair-wise correlation of the five deleteriousness scores in the (a) disease-causal and (b) neutral rare variant sets. The Spearman's rank correlation method was used to calculate the pair-wise correlation coefficients.

Figure 4.

Receiver operating characteristic (ROC) curves of various methods. (a) The control (neutral) variants are rare (MAF < 0.01); (b) the control (neutral) variants have MAF ≥ 0.01. The true positive rate (sensitivity) and false positive rate (1-specificity) of logistic model were obtained by 10-fold cross validation procedure. Logistic model: performance of conventional multiple logistic regression model when combining the five deleterious scores (PhyloP, SIFT, PolyPhen2, LRT and MutationTaster). The true positive rate and false positive rate of the five different prediction methods were generated by varying the threshold scores for prediction in the entire data set.