The first reason I recently suggested for why biotech ML projects often get stuck at the proof-of-concept phase was when the problem you’re solving can’t be acted on. The fact is, it’s not hard to find problems that the scientists are interested in answering. The problem is when you provide the answer and they don’t know what to do with it. So how can you use the hammers and nails conversation to figure out which answers will have the most impact?
The framing that I’ve found most useful for this is to think of each step of analysis as helping to define the next experiment. I know, I know. This seems like a very narrow conception of what AI/ML should be capable of. But this isn’t for marketing - you can keep this framing to yourself.
But think about it - No matter how accurate your model is, it’s always going to be followed by validation in the lab. That’s just how it works. So if you can figure out how to translate your predictions into a validation experiment, you know exactly how the bench team can act on it. If that next experiment discovers something that moves the program forward (whether or not it’s a direct validation of your prediction) then that’s tangible impact.
If, on the other hand, you can’t translate your prediction into a next experiment, maybe you should pick a different nail.
Scaling Biotech is brought to you by Merelogic. We’ll help you turn your ML prototypes into tangible impact, whether it takes a few small tweaks to how your team operates or larger changes to your tools, infrastructure and projects. If you want to explore what this might look like for your team, send me an email at jesse@merelogic.net
Wait, are you suggesting that the outcomes should be actionable? That's can't be true...