Feed post
Jérémy Armetta likes this
Senior Scientist, Automation at Novonesis
9h •
Follow
When it comes to getting AI agents and automated science to generate useful outcomes, I believe that (one of) the devils will be in the metadata of previous experiments, or lack of.
If you’ve ever dealt with liquid classes in any liquid handler software, you know that they are not trivial to extract and compare across systems. For your data scientists working on your LIMS and AI platform, there is no chance that they will be able to figure out that the reason they get different results out of two same-system liquid handlers, might be due to slight differences in the liquid classes used on the two machines – at least in early R&D. Clinical may be running a tighter ship in that regard! 😉
What I've come to like about #PyLabRobot, is that liquid classes (and everything else) is prepared in a simple tabular form. This makes it easier for both wet lab and data scientists alike, to keep track of -how- the liquid handler actually transferred the samples and solutions: Across all workflows, across all labs.
Making liquid classes, and other aspects of the automated workflow explicit this way, should make it a lot easier to ID the subtle, underlying, differences that can cause variation in your experimental results, and in that way give your AI strategy a better foundation to evolve out from! 🤖
Is there any experimental metadata that you feel like is not getting the attention it need in order to succeed with AI and other automated science approaches? 🙂
1,000