Effect of influential observation in genomic prediction using LASSO diagnostic

Abstract

Detection of influential observation is one of the crucial steps of pre-processing to identify suspicious elements of data that may be due to error or some other unknown source. Several statistical measures are developed for detection of influential observation but still challenges are there to detect a true influential observation for high dimension data like gene expression, genotyping data. In this article we have demonstrated the effect of influential observation on genomic prediction accuracy by using recently proposed LASSO diagnostic, i.e. Df-Model, Df-Regpath, Df-Cvpath, Df-Lambda and Influence-LASSO. The effect of influential observation on genomic prediction accuracy was explored by observing the change in estimated and true accuracies for dataset with and without influential observation scenario. For this purpose we have used wheat and maize datasets which are available in public domain. It has been observed that influential observation had significant effects on the genomic prediction accuracy. In this study it has been shown that by implementing efficient diagnostic measure for influential observation detection, accuracy of genomic prediction can be improved.