The team behind DeepMind’s protein structure prediction software AlphaFold announced Oct. 31 that it is expanding its model beyond proteins. Ligands, nucleic acids, and post-translational modifications will all be included in the update, says the team in a blog post and accompanying white paper.
The announcement comes hot on the heels of the preprint describing a similar expansion to the RoseTTAFold model built by David Baker’s group at the University of Washington. Neither team has yet made their new software or code available to the broader scientific community.
Modeling biomolecules, and the chemicals that can interact with them, has been a huge topic of research interest in recent years. One key application is in drug discovery, with established pharma companies and start-ups investing in the area. For example, in 2021, DeepMind CEO and cofounder Demis Hassabis launched Isomorphic Labs, which will use the new AlphaFold model in its work to design new medicines.
But the rapid developments in computational protein structure modeling and design have alarmed some researchers. Just as these new computational approaches could speed up drug and vaccine design, so too could they be used for more harmful applications. On Oct. 25, the Institute for Protein Design hosted an AI safety summit where it was agreed to develop new voluntary guidelines for researchers using AI for protein research. The UK’s AI Safety Summit, hosted just outside London, started Nov. 1 and will discuss the threat of AI systems helping create new chemical or biological weapons.