A.I. for Drug Development

Deep Linking
March 10, 2020
Prioritizing Accountability and Trust
March 10, 2020

A.I. for Drug Development

In 2018 and 2019, drug companies ramped up research to determine if A.I. could be used at every phase of drug development.

In 2018 and 2019, drug companies ramped up research to determine if A.I. could be used at every phase of drug development, from hypotheses, picking better compounds and identifying better drug targets to designing more successful clinical trials and tracking real-world outcomes.

Microsoft and Novartis announced a collaboration for A.I.-driven drug discovery, Pfizer intends to use IBM’s Watson, and Alphabet’s DeepMind proved last year how a tech company could beat a roomful of biologists in predicting the shape of a protein based on its genetic code.

Nearly every major pharmaceutical company inked deals with A.I. drug discovery startups, too, including Johnson & Johnson, Novartis, Merck, AstraZeneca and GlaxoSmithKline. And investors poured $2.4 billion into hundreds of such startups between 2013 and 2019, according to data by PitchBook.

Much of the potential in A.I. stems from deep learning’s ability to sort through huge volumes of information and to learn and extrapolate from that information.

The upshot: A.I. can think faster than humans—sorting data in months versus years—and see patterns that we may not. Still, drug discovery is tricky, because the algorithms rely on drug targets that must be published in research journals and have well-characterized metabolic mechanisms. Most data about potential compounds, too, isn’t always readily available, and when it is, it isn’t always complete or reliable.

Filling in the gaps and cleaning up that data takes time and money. It also requires data sharing—and most drug data is proprietary and locked up by big drug makers. Despite the frenzy in the industry about the technology’s potential, no A.I.-driven drugs have been created yet. But the industry may be inching closer: In September of 2019, Deep Genomics in Canada successfully used A.I. to decipher more precisely how one gene mutation fails to create a protein, one of hundreds of genes that leads to Wilson’s disease, a life-threatening genetic disorder in which the body’s ability to properly distribute copper is impaired.

Deep Genomics used another set of algorithms to analyze billions of molecules and ultimately identify 12 drug candidates, which appeared to work in both cell models and mice. The company will take one of them, known as DG12P1, to human clinical trials in 2021. The process took 18 months, instead of the traditional 3 to 6 years. If A.I. works for drug development, it will dramatically alter the field’s needed skills in the future: drug developers must not only know biology but computer science and statistics, too.

Then there’s the Food and Drug Administration approval process. Using algorithms for drug development brings up a host of ethical questions. Will bias invade drug discovery much like it has other arenas of A.I., thereby marginalizing certain patients or diseases? Do algorithms need their own clinical trials? Could A.I. be used to take shortcuts and undermine the value of the science being done inside the laboratory? Advocates say A.I. will make drug development and clinical trials more efficient, thereby cutting drug prices and paving the way for more personalized medicine.