Increase your training set with real, freely available data!

Andrew’s paper shows that homologs can be used to augment sequence data