Simulation of plant complex trait architectures using deep neural networks

Abstract

Although the use of molecular markers SNPs (single nucleotide polymorphisms) hasmade a major contribution to genetic improvement programs, their application is still limited in the case of sugarcane, a species with high genomic complexity. Due to the lackof biocomputational tools compatible with this species uniqueness, specific methodologieshave to be used. As a result of sugarcane immeasurable importance to the biofuels andsugar manufacturing sectors, breeding of economically important cultivars makes a majorcontribution to the world economy. Genomic selection methodologies have proven to bean alternative to traditional marker-assisted approaches because of the polygenic nature ofquantitative phenotypic characteristics and the consequent complexity of applying methodsfor genotype-phenotype association. This research project proposes an innovative approach for sugarcane selection. Using a matrix of putative SNPs identified on Genotyping bySequencing data and softwares adapted to the species polyploidy and aneuploidy, an association modeling between these markers will be performed using complex networks. Throughcommunity detection methods in the built network, it is intended to perform data imputation, dimensionality reduction and identification of SNPs with related biological categories.To build the selection model, machine learning techniques will be used as nonlinear models,modelled to capture the specificity of the sugarcane genome. (AU)