A British startup Exscientia claims it has developed the first medication created using artificial intelligence that will be clinically tested on humans. The medication, which is meant to treat obsessive-compulsive disorder, took less than a year from conception to trial-ready capsule.
How was AI used? There are lots and lots of possible molecules that might be useful in medications, far too many for all the medical researchers in the world to manually test. But by using different types of AI, a computer system can come up with and mine through different molecules, comparing them against different parameters and learning the most promising compounds faster than a human could.
https://www.vox.com/2020/1/31/21117102/artificial-intelligence-drug-discovery-exscientia
Carrying on with my personal theme of skepticism is healthy. The AI component to this is “Centaur Chemistry” - I cannot find any academic reference to what this is. The use of various computation approaches to library development, lead optimization, etc, have been around for at least 15 years (I took whole seminars on QSAR - quantitative structural-activity relationship modeling - while in training in clinical pharmacology in the US). The descriptions of how this specific companies approach is better or worse than existing approaches is fuzzy and hyped at best in the literature. This appears to be about lead optimization. It remains very unclear how they could have sufficient pre-clinical safety data to actually get a novel concept into people in that time frame - e.g. I can find nothing published about their animal safety studies in the literature (which often includes chronic tox studies in the most susceptible species that are 12-18 month studies). There is something quite fishy about this story in general, and I suspect a lot of this is hype. Plus, the failure rate for psychiatric drugs from pre-clinical to marketed drug is on the order of 90+%, and that part of the pathway is presently 7-10 years. I would be interested if they can get something from nothing to approved drug in < 5 years. I would be way more interested if they can do it twice.
That said - there is a lot of interest in these tools in drug development - although it is interesting that major pharma recently had a distinct cooling off of their exuberance for these methods.
I’ve attached a pretty good overview of the ML in drug development. It does finish with the caveat “Finally, the use of statistical learning algorithms is not short of challenges and should be handled with care”
uploadedfile:335849668608, opengraphobject:[360518153117696 : https://www.vox.com/2020/1/31/21117102/artificial-intelligence-drug-discovery-exscientia : title=“Would you take a drug discovered by artificial intelligence?” : description=“An OCD drug created via AI will be tested on humans.”]
Thanks for your insights and additional information Matthew! Yes, this one read as having quite a bit of hype behind it, especially with the timeframes given for potential (and as you said with current 90-ish % failure rate for drugs from pre-clinical to market) availability to the public.
It’s definitely not my area of expertise, so I appreciate others sharing their thoughts and having good discussions - and in turn educating me on things that are a wee bit outside of my day-to-day!
Agree with the final caveat too.
opengraphobject:[360518153117696 : https://www.vox.com/2020/1/31/21117102/artificial-intelligence-drug-discovery-exscientia : title=“Would you take a drug discovered by artificial intelligence?” : description=“An OCD drug created via AI will be tested on humans.”]
This seems like a really good example of drug discovery with an element of AI. The actual article does a great job of emphasizing the actual role AI played in this, is not hyped, and emphasizes the huge amount of work that additionally went into the preclinical discovery phase. In brief: trained a openly available neural network that is well referenced and code available, to classify manually binarized library of chemical structures as being good antibiotics or not. then re-applied that trained NN to classify a new chemical library that was too large for humans to manually review. used pre-set thresholds, and reasonable business logic, to select the molecules that were both most likely to effective, and already in Phase I/II clinical development. THIS point is critical - because by repurposing a drug in development, they may cut out literally 10 years and a billion dollars in R&D. Did the highly labor intensive and detailed science to validate their hit in the lab.
uploadedfile:358526754816, opengraphobject:[360518153117696 : https://www.vox.com/2020/1/31/21117102/artificial-intelligence-drug-discovery-exscientia : title=“Would you take a drug discovered by artificial intelligence?” : description=“An OCD drug created via AI will be tested on humans.”]