Identification of relevant subtypes via preweighted sparse clustering