This manuscript is a community perspective about recommendations for protein engineering in the low N setting [@doi:10.1038/s41592-021-01100-y] where there is little experimental sequence-function data available. It focuses on techniques guided by machine learning.