GAPIT3: New Architecture to Quickly Adapt Third Party Packages for Genomic Association and Prediction

Maintaining and development of software packages for genome wide association studies (GWAS) are challenging not only because the demands of problem-free application, but also because that computing speed must be boosted for euphonically growing data size and statistical models must be advanced to control both false positives and false negatives. GAPIT is one of the software packages widely used for GWAS and genomic prediction. Several handy features in GAPIT were widely used, including Principal Component Analyses (PCA), kinship, data visualization (e.g. population structure in 3D) and results interpretation (e.g. Manhattan plot and QQ plot). The package implemented multiple statistical methods to improve speed and statistical power, including the compressed mixed linear model (CMLM) developed in 2010, enriched CMLM and SUPER methods developed in 2014. Multiple methods and packages were developed to produce better GWAS results, including MLMM and FarmCPU. There is critical need to develop a package that can quickly integrate existing and new developments. We redesign GAPIT to fit the need. The new version of GAPIT (version 3) is capable to adapt third party packages easily. Here we describe the design and implementations for adapting original version of GAPIT and the most recently developed packages, MLMM and FarmCPU. The advantages and the performances of using these packages in GAPIT3 were compared to using them along. GAPIT3 has been released to public and GAPIT users has started to use the new version of GAPIT substantially.