Use identity and similarity to find nearest germlines

Previously we used number of mutations, which caused long J genes to be discarded even if most of the sequence matched better.