EY: What impact could AI and related machine learning tools have on the speed and cost of drug R&D?
Jackie Hunter: Artificial intelligence has the potential to impact the whole drug discovery and development process. As an industry, we're still losing 50% of compounds in Phase II and Phase III trials for lack of efficacy. That isn’t sustainable; it tells us we're picking the wrong targets. A further quarter of failures in Phase II or III are for strategic or commercial reasons. That also tells us industry is not always making the right decisions about what compounds to prioritize.
Our deep-learning platform could lead to a fourfold increase in R&D success rates up to and including target validation. We already have some evidence for that: in less than a year, we have generated 36 new hypotheses and validated 24 of them in vitro. Traditional biopharma R&D would typically only manage about five in that timeframe with the same personnel.
We're also using our deep-learning supercomputer to generate chemistry models in less than a week, rather than a couple of months. It remains to be seen whether this acceleration translates to clinical proof of concept and beyond. But it's exciting.
EY: How does BenevolentBio’s AI platform work, and what kinds of insights does it generate?
Hunter: The system ingests all kinds of scientific information — public, private, structured, unstructured — and annotates it with specialist biomedical dictionaries. Then we apply natural language processing and other algorithms to build a knowledge graph, showing the complex pattern of interactions between various molecular entities and diseases. This allows us to generate new potential associations or rule out existing hypotheses. Negative associations are sometimes even more valuable than positive ones, in terms of decisions to discontinue a particular approach.