When do I use AI and why?

More frequently than you would guess

As I mentioned previously, AI shows promise for therapeutic drug discovery in terms of ligand discovery and design, but will have challenges related to predicting clinical trial results and indications.

My lab uses AI-esque approaches continuously as follows:

• Ultra-large-scale docking (now with Trillions of molecules) of theoretical compounds against theoretical structures

• We predict molecular targets for small molecules

• We use AI (Topaz Train) to solve and optimize structures

• I use GROK daily for literature reviews, although I also do my own lit reviews as the output is typically incomplete

• I recommend to my lab mates to use their favorite AI platform to ‘find the best antibody for protein X’: this is particularly useful as you can ask it to find the number of papers which cite each particular antibody and then you can look them up.

My verdict is that AI is very, very useful for certain routine tasks (lit reviews, computationally intensive projects).

AI fails when there is no reliable training and validation data

So yes, we use AI-esque approaches all the time—for structural biology and the initial phases of drug discovery they really speed things up. Once you open up the biology hood, though, things become complicated quickly.