Nuformix: AI can't take you where we go

Nuformix’s business model requires that it investigates possibilities where drugs already in use can, using cocrystal technology, be reprofiled for novel treatments. Clinical studies for the company’s lead asset, NXP001, a treatment for chemotherapy-induced nausea and vomiting, have just started. There are five other cocrystals in the portfolio, and the search is on for more.

Conventional wisdom holds that the search for new drugs to reengineer can be done by AI, but Nuformix CEO Dan Gooding sees it differently.

"Big data is still too conceptual for Nuformix to easily integrate into our R&D," says Dan. "AI can assist, but in our experience just augments the process - you get to the same answer a bit faster. I find AI still quite untangible - and I say this after experiences applying AI to our discovery and development: it's a black box from which we’re yet to see delivery of something a human couldn’t do.

"The world of global research is vast. Computation can assist, but AI results still need to be checked that they make sense. What’s key to us in ideation - regardless of the source - is validation of rationale and practical experience. Interaction with our network and clinicians is key - understanding patient needs at the coalface. That's anecdotal observations and knowledge that can’t be gleaned by AI or anything else. Perhaps a clinician somewhere has made a connection between a new disease target and a known drug and decides to test that theory, perhaps in a human or an animal and publishes their results - increasingly common in literature. If positive, we might be in a position to make that drug so much more useful than it was - that's what we put into the mix, that's what we're scouring for - validation of an innovative therapy where we can add value. AI can sometimes point you to more ideas faster, but the rationale still needs to be challenged by a human.

Ultimately there's no substitute for real data. What do AI algorithms do? On the whole they're looking for contextual links to what a drug might do, but the algorithm is only as good as the context you give it. We need quality - small numbers of well validated concepts, not quantity."

“For example,” says Jo, “with NX002, I was looking for a completely different indication, if it was an algorithm it wouldn’t have gone off piste in the way that I did.

“AI will get there but if you’re asking for a certain thing it won’t spot the other thing that you weren’t looking for, but it’s there. You can spot it but the algorithms won’t.”