the catch with a genome-wide map is the benchmark labels came from conservation, so you can't tell prediction from re-reading the prior.