Articles
Using machine learning for the discovery of epistatic fruit rot resistance markers
Article number
1440_11
Pages
79 – 84
Language
English
Abstract
Marker assisted selection (MAS) is an invaluable tool utilized in crop breeding for selection of genotypes and genomic selection (GS). Currently, most methodologies rely on markers developed from either quantitative trait loci (QTL) studies or genome wide association studies (GWAS). However, many phenotypes are not limited to single or few loci easily identified in QTL and GWAS methods, but often are the result of several small and interacting (e.g., epistatic), variants of multiple loci.
Machine learning (ML) methods offer an alternative means of providing informative markers for GS with an increased ability to identify interacting and epistatic markers.
We developed ML methodology to study the polygenetic trait of fruit rot resistance in cranberry, where identifying single QTLs with a reliable marker has been difficult.
Our approach utilized cranberry accessions previously found to be resistant to field fruit rot, yet were not well characterized using conventional QTL analysis software.
Six populations were selected for phenotyping and GBS analysis.
Marker selection was performed through a ML approach using random forest regression.
Random forest regression and GBS sufficiently identified four major epistatic loci on four different chromosomes with three loci contributing to an estimated 23% of phenotypic variance in one population and two loci contributing to an estimated 27% of the variance in the remaining population.
Machine learning (ML) methods offer an alternative means of providing informative markers for GS with an increased ability to identify interacting and epistatic markers.
We developed ML methodology to study the polygenetic trait of fruit rot resistance in cranberry, where identifying single QTLs with a reliable marker has been difficult.
Our approach utilized cranberry accessions previously found to be resistant to field fruit rot, yet were not well characterized using conventional QTL analysis software.
Six populations were selected for phenotyping and GBS analysis.
Marker selection was performed through a ML approach using random forest regression.
Random forest regression and GBS sufficiently identified four major epistatic loci on four different chromosomes with three loci contributing to an estimated 23% of phenotypic variance in one population and two loci contributing to an estimated 27% of the variance in the remaining population.
Publication
Authors
J. Kawash, I. Dehzangi, J. Polashock
Keywords
QTL, GBS, machine learning, epistasis, boosted regression
Online Articles (74)