AI that finds disease targets and predicts a drug’s chances of FDA approval

This is such a strange article. Immediately criticizing the FDA’s approval process as invalid and time wasting without citing exactly why. I’m sure there are limits to the existing process, so what are they, and what is it doing right? They also mentioned several times that an AI has found a means to correlate gene expressions with drug interactions…and that’s it. Just pasting the abstract isn’t the neutral reporting they think it is.

What would be really interesting is if doctors had genetic profiles of their patients and could analyze side effect probabilities specific to you, rather than a group in the case study when the drug was approved. Even having this anonymous information available to the drug research could help simulate potential avenues.

The AI discovered an incidental proxy of FDA success - measuring funds level of the offshore banks of the comittee that approves a drug.

It would be incredible--in the literal sense of the word--if this actually works out-of-sample.

The biology itself is mind-bendingly hard. On top of that, FDA approval depends on all sorts of factors beyond "efficacy": severity of the disease vs severity of adverse effects, other options, and0--lately--a decent amount of capriciousness.

Here's the actual article (OA/CC-BY), which is a bit less...expansive.

It predicted A FEW FDA-approved outcomes; it does not take arbitrary predictions of any disease from target to approval.

Sahoo, D., Swanson, L., Sayed, I.M. et al. Artificial intelligence guided discovery of a barrier-protective therapy in inflammatory bowel disease. Nat Commun 12, 4246 (2021). https://doi.org/10.1038/s41467-021-24470-5 or https://www.nature.com/articles/s41467-021-24470-5

My first take on the headline was that AI found a disease that targets and predicts a drug's chances at FDA approval. That was like !!WOW!! until I reread the title correctly.