I think a very good thing to do is to write test code (almost never done in my experience) with made-up datasets where you know exactly what the results should look like.
I think general programming best practices would apply, particularly code review and planning your code and breaking it up into discrete parts. I have seen many scripts for research that are 5k LOC of spaghetti that become impossible to read or adjust.
We're talking about scientists, not computer scientists. As surely as business managers and industrial engineers don't need to know the best programming practices to be good SQL monkeys, surely scientists- neuroscientists at that- don't need to know general programming practices.