It is difficult to grasp the pace of scientific progress these days. One benchmark is how we do our work. The most common programs people use to carry out their analyses were unavailable when I started my PhD research, and many of the key methods had not been invented yet. This could be because we're in a particularly fruitful time for research, but I tend to think it's more of a sign of things to come. We need to be prepared for the fact that the next batch of scientists, 5-10 years from now, will be applying techniques that have not even been invented yet on data sets that we can hardly imagine. This perspective is expressed well by Ken Robinson. (Thanks to Larry Forney for pointing me to that video).
What does this mean? To me, it suggests that there's a serious lack in training of graduate students. In other math-intensive fields (like physics), students are required to take a variety of math courses to prepare them to deal with complex data and equations. In biology, students sometimes take these courses, but it's usually not required. I think it's a key ingredient for success in an uncertain and fast-moving future.
What classes are the most valuable? To me, these have had the most pay-off:
1. Probability (something more advanced than a basic stats course)
3. Matrix algebra
Take a math course or two, it won't kill you.