Discovering Intermediates in Biosynthetic Pathways that have Interesting Properties

An interesting take from one of the most brilliant minds of microbiology, Itzhak Mizrahi. I love the paper he is highlighting, the results are impressive and exciting, it took a lot of "fundamental microbiology and genetic engineering" to get to it: the discovery of intermediates in methylenomycin biosynthesis (premethylenomycin C and its lactone precursor) that are even more potent than methylenomycin A against strains of Gram-positive bacteria including Staphylococcus aureus and Enterococcus faecium.

But I have to disagree with the (implicit) premise that AI has no role to play in such discoveries. While I admire the human effort and agree that there is no shortcut to the lab work (at least in 2025), I also want to point out that the study really began in 2006, with a complete biosynthetic map by 2010, and it took until 2017 for a PhD student on the team to test intermediate molecules for antimicrobial activity. According to a Nature interview (link), Greg Challis said “We discovered these intermediates, and we left them for a while because we didn’t quite know what to do with them.” In other words, it is a beautiful example of serendipity in science combined with hard work and persistence.

And that's where I think AI can play a significant role, by augmenting serendipity, that is, by turbocharging the generation of relevant hypotheses.

To illustrate this point, I used chatGPT-5 Thinking (tried to run FutureHouse, my favorite bio-AI but it will be back online in 1 week) by limiting literature references to up to 2022 with a very naive question about which intermediates in methylenomycin biosynthesis might have interesting and useful properties. It came up with a bunch of interesting ideas with ways to test them, too long to all list here, DM me if you'd like the full, detailed output -I also explore potential mechanisms for antimicrobial activity:

"There are chemically credible reasons to expect some late mmy intermediates to show useful properties, especially anti-Gram-positive activity, adjuvancy (permeability/efflux), biofilm interference, and regulatory probing of TetR-type circuits."

In other words, AI can give you a 7-year boost in generating useful hypotheses. This is obviously generalizable to any biosynthetic pathway, especially once the intermediates have been identified.

Andrew White, Samuel G. Rodriques, Jordan Shlain, MD, John Battelle, Kevin Weil, Mike Krieger