Using a Random Forest proximity measure for variable importance stratification in genotypic data

In this work we study variable-significance in classification using the Random Forest proximity matrix and local Importance matrix. We use the prox- imity m atrix t o g roup t he s amples acr oss a num ber of c lusters a nd use t hese clusters to s tratify th e importance of a v ariable. W e apply t his a pproach t o a cardiovascular g enotype d ataset f or sample classification b ased o n coronary heart disease and we found a number of variations related with cardiovascular disease phenotypes. We also used a set of phenotypes related with this genotype data to match the obtained clusters with coronary heart diseases phenotypes.