Six things to keep in mind while reading biology ML papers

What’s interesting is that a lot of bio ML papers claim they can make new discoveries using their methods, but never actually make any new discoveries. Even high-end papers would stop at rediscovering some known biology rather than find something previously unknown.

Bio ML papers can spit out all sorts of predictions, but it’s hard to distinguish the nonesense with the real, and there is no sense of the false discovery rates.

If you want to convince non ML people about your method, go discover something new, unexpected, or surprising. You will convince people much more than gaming benchmarks.