Genome wide biological markers and DNA microarrays are the most used biological signals to identify the association with diseases and traits of interest. The aim of this work is to develop new algorithms in the area of association study using genetic markers and the area of missing values estimation for DNA microarrays. Kernel modelling techniques are proposed in both areas, based on the assumption that biological signals are generated from complex interactions between genes and environmental factors. Genetic markers dataset is an abundant resource of DNA variations in a population. A character of genetic markers dataset is the high number of dimensions of the markers but limited number of samples. The algorithm developed in this work uses a two-stage structure to predict the trait of milk protein yield by using the genetic markers of dairy cattle. The first stage of the algorithm selects a subset of the markers related to the trait of interest. The selected sparse markers are fit into the kernel models in the second stage for the prediction of the trait.