If we run prodigal on small genomes, it is not possible to train (if <2000nuc). So, instead, just run prodigal, without a training file, using its '-p meta' procedure