I think

likelihoodType

picks a statistical model, not a biological label — ONCO expects clustered recurrent variants, TSG expects dispersed inactivation plus biallelic loss.