Google DeepMind: Drug developers seek a structural advantage from AI

Receive free Lex updates

Dr Google gets a bad rap. Researching symptoms online makes cyberchondriacs sick with worry. But it would be hard to snipe at the Alphabet-owned business’s latest contribution to the diagnosis of disease. Google DeepMind’s new artificial intelligence tool predicts whether mutations in human genes are likely to be harmful. That should speed up the detection of diseases caused by rare genetic variants.

The achievement demonstrates the power of AlphaFold, the protein-shape-predicting software on which it is built. Since AlphaFold’s predictions became freely available to researchers in 2020, they have been enthusiastically adopted. The admittedly biased chief executive Demis Hassabis reckons AlphaFold has “unequivocally [the] biggest beneficial effects so far in AI on the world”. 

Big Pharma bosses hope such tools can improve its feeble productivity. Development costs are estimated at $2.3bn per drug by Deloitte, leaving return on R&D investment at a pitiful 1.2 per cent. AI should improve matters by speeding up drug discovery. Morgan Stanley says AI could cut pre-clinical development costs by as much as two-fifths. It could create a $50bn market over the next decade.

More than 200 start-ups are competing for a share of the market, according to CB Insights. DeepMind’s own drug discovery start-up Isomorphic Labs is one. Despite the VC slowdown, there has been a stream of recent deals. Germany’s BioNTech recently acquired UK-based InstaDeep for $682mn. Eli Lilly inked a $250mn deal with Shenzhen-based Xtalpi in May. In July, Nvidia invested $50mn in US-based Recursion.

Using AI does not guarantee success. Clinical trials of the first AI-designed molecule — announced in 2020 by Oxford-based Exscientia and Sumitomo Pharma — were unsuccessful. In May, London-based BenevolentAI announced it was laying off 180 staff, after its lead drug candidate failed.

Nonetheless, Big Pharma’s determination to exploit its potential is encouraging. Given that nine out of 10 new drugs fail, there is huge scope for improvement.