I did AI for drug discovery in the mid-90s. Most of the work was data cleaning. J. Med. Chem. papers routinely included impossible structures. How believable were their assay values then? Cc @curiouswavefn
-
-
Show this thread
-
I got a lot of flak from my boss. “We hired you to do AI, not spend all your time fixing published work!” The fantasy that fairy dust can somehow save you from the hard work of figuring out what data mean is perennial.
Show this thread
End of conversation
New conversation -
-
-
All true… at the time, we were the only ones using neural networks, or a sane shape representation, so differentiation wasn’t an issue. And we did have some proprietary data sets—but their quality was probably worse than the academic literature even.https://pubs.acs.org/doi/abs/10.1021/jm00041a010 …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Indeed. So now imagine building a search tool that lets you search and visualize this data fast. It means you’ll now be able to find - and use - bad data much faster than before.
New conversation -
-
-
In one of my projects there was a cyclized side product causing potent inhibition. Took a lot of forensic work to figure it out.
-
Science is like this in every field, afaict. [thread:]https://twitter.com/Meaningness/status/1129952518841163778 …
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.