The UK-based AI drug discovery company, Exscientia, has been hailed as reaching a huge milestone: the world’s first AI-designed drug to enter Phase 1 clinical trials. Why is this a big deal? And are computer-designed drugs a good thing?
Exscientia’s announcement is a big deal for many reasons. It is a sign that the promised synergy between computer power and drug discovery is beginning to bear fruit. Early-stage drug discovery used to be a trial and error process guided by the experience and knowledge of talented researchers, but the months of lab time can be replaced by minutes of analysis by a series of algorithms reviewing electronic databases of virtual compounds. There are an estimated 1060 potential drug compounds, so AI can make a big difference in finding the most promising compounds.
In the case of Exscientia’s compound, code-named DSP-1181, using AI saved a massive amount of time and effort. To find a compound suitable for Phase 1 clinical trials often takes 5 years and the review of 2,500 other potential compounds. Exscientia achieved this in 1 year and after reviewing 350 compounds.
But there are some concerns about putting drug discovery in the metaphorical hands of computers. Computers are only as good as their programmed instructions and turning the experience of a team of drug discovery researchers into such instructions is a huge challenge. Even with the help of machine learning, the programme is only as good as its input data. In practice, drug candidates discovered by computers are often plagued by problems that make them unworkable in reality.
Exscientia gets around this by keeping a human in the loop. Even better, they keep world-class chemists in the loop, specifically partner organisations like Roche, GSK and Sanofi. Their candidate DSP-1181 was developed in partnership with Sumitomo Dainippon Pharma Co., bringing together Sumitomo’s expertise in the psychiatry and neurology field with Exscientia’s expertise in computational chemistry.
Combining the best of human and machine is a powerful combination. Indeed, this approach is being proposed in other medical applications of AI, for example in spotting breast cancer on mammograms. DSP-1181’s entry into clinical trials is an exciting step for AI-aided drug discovery, but (for now), a human will remain in the loop and the best use of this new drug discovery technology will be where the tech teams and chemistry teams work together to play to their strengths.