Welcome to GenEpi’s docs!

GenEpi[1] is a package to uncover epistasis associated with phenotypes by a machine learning approach, developed by Yu-Chuan Chang at c4Lab of National Taiwan University and AILabs.

Cannot link to image

The architecture and modules of GenEpi.

Introduction

GenEpi is designed to group SNPs by a set of loci in the gnome. For examples, a locus could be a gene. In other words, we use gene boundaries to group SNPs. A locus can be generalized to any particular regions in the genome, e.g. promoters, enhancers, etc. GenEpi first considers the genetic variants within a particular region as features in the first stage, because it is believed that SNPs within a functional region might have a higher chance to interact with each other and to influence molecular functions.

GenEpi adopts two-element combinatorial encoding when producing features and models them by L1-regularized regression with stability selection In the first stage (STAGE 1) of GenEpi, the genotype features from each single gene will be combinatorically encoded and modeled independently by L1-regularized regression with stability selection. In this way, we can estimate the prediction performance of each gene and detect within-gene epistasis with a low false positive rate. In the second stage (STAGE 2), both of the individual SNP and the within-gene epistasis features selected by STAGE 1 are pooled together to generate cross-gene epistasis features, and modeled again by L1-regularized regression with stability selection as STAGE 1. Finally, the user can combine the selected genetic features with environmental factors such as clinical features to build the final prediction models.