ISMU 2.0: A Multi-Algorithm Pipeline for Genomic Selection

Genomic selection (GS) is getting immense popularity in modern breeding especially for complex traits. In the GS, marker effects, after deploying genome-wide and high-density marker datasets, are estimated without testing for their statistical significance and used to predict genomic breeding values (GEBVs). Although some groups have developed algorithms for calculating GEBVs, these algorithms are often coded in R, FORTRAN and C++ languages that require understanding of programing environment and large-scale data handing. In addition, breeding community may have a challenge in applying several algorithms together due to their limited expertise in different algorithms. To overcome these challenges, as part of the Integrated Breeding Platform, ICRISAT and its partners are in process of developing “ISMU 2.0” pipeline after integrating several GS algorithms in Java. For instance, the pipeline incorporates seven popular GS algorithms namely RR-BLUP, Kinship Gauss, Bayesian LASSO, BayesA, BayesB, BayesCπ and Random Forest written in FORTAN and R languages. RR-BLUP and BayesA algorithms implemented in FORTRAN language in the ISMU have been found faster in terms of analyzing large datasets. Additionally, the pipeline has modules to check quality of marker datasets such as % missing data, minor allele frequency (MAF) and polymorphism information content (PIC). The pipeline has been optimized for running MS-Windows, Linux, CentOS and Ubuntu platforms.