An ML drug discovery startup trying really, really hard to not cheat owlposting.com/p/an-ml-drug-… on the 12-person, Utah-based startup @leashbio, their culture of rigor, and the many ways small molecule models accidentally learn the wrong thing
The ML drug discovery startup trying really, really hard to not cheat (Leash Bio)

The ML drug discovery startup trying really, really hard to not cheat (Leash Bio)

6k words, 27 minutes reading time

owlposting.com

Dec 23, 2025 · 1:35 PM UTC

8
18
11
168
50,284
RelevantRecentLikes
update: Leash is actually not a 12 person startup, rather 9 person :)
1
6
1,249
The 'Money Stuff'-style introduction was excellent, I'm a big fan. (Though I don't know if that was an explicit inspiration or not!)
1
4
1,327
haha, i have never actually read any of matt levines stuff that had a similar introduction, would love to see it!
1
625
This is an epic post. Would love to go for drinks in San Fran during JPM if you’re about! Merry Christmas!
1
1
238
would love to!
207
sounds like they’re in a constant battle with their own algorithms
320
It’s the only way anyone should practice science. The replication rate is abysmal, if it does replicate the interpretation is often wrong, and if we get that right, the result’s magnitude is often trivial or it doesn’t generalize. Generously, 1 of 1000 papers are publishing.
1
1
168
Really happy to see leash working on this and great article! But there are lots of confounders in DEL data too. From incomplete synthesis to effect of linker to dna positioning etc, even the binary signal might not represent a true binder but rather a mixture of various effects.
1
1
153
They drafted a 6k-word article on a problem already well-known in the field as the dereplication problem. That’s why you need to become a subject matter expert before jumping into VS Code and “discovering” the next drug.
154