Development of temporal genomic selection models via Bayesian Networks applied to sorghum bicolor

Abstract

New Genomic Selection models that allows predicting sorghum (Sorghum bicolor) traits early in season using repeated measures data obtained from robotic platforms may help to increase predictability accuracy of genotypes after successive circles of selection using only molecular markers information. However, to date, there is no developed Genomic selection models that can recovery temporal genetic information and conjointly are computationally suitable to be applied in scenarios Big Data. Bayesian inference in conjunction with Bayesian networks offers high versatility to model complex natural processes. Using these modeling frameworks we propose to develop new statistical models to deal with this situation. We are going to analyze a unique outstanding database, which includes phenotypic and genomic data collected by high-throughput platforms. Phenotypic data will be collected by a robotic platform under development called TERRA-MEPP. Genomic functional data will be collected with the most advanced platforms like second-generation sequencing, sequencing with reduced-representation of the genome and DNS-chip profiling. We expect more than <180 millions repeated measures phenotypic data collected until the end of the season and approximately <300,000 functional variants (SNPs) discovered. This dataset will be provided by Dr. Michael Gore from Cornell University (USA) who will be out partner in developing the models and analyzing/interpreting the results. We are going to propose, validate and test our Genomic Selection models via Bayesian Networks. (AU)