Variable Selection from Random Forests: Application to Gene Expression Data