NULL protein IDs would give query errors because the field can't
 be NULL, and transcripts not mappable to the current build would
 throw notices.