An AI answers one of biology's biggest problems
Google's DeepMind this week reported solving one of biology's long-standing problems: predicting the 3D structure of proteins.
Why it matters: Being able to determine protein structure could help to speed up drug development and aid researchers in understanding the basic biology of disease.
- “It’s a game changer,” Andrei Lupas of the Max Planck Institute for Developmental Biology told Nature News' Ewen Callaway.
The problem: Predicting how physics arranges the atoms in amino acids, giving rise to the twisted and folded structure of proteins, is one of biology's toughest challenges.
- A protein's structure determines whether and how it binds to other proteins and molecules — biological processes that underpin life.
- The structure also plays a role in how drugs bind to proteins in the body.
- “We have been stuck on this one problem — how do proteins fold up — for nearly 50 years," John Moult, a professor at the University of Maryland and a co-founder of the Critical Assessment of Structure Prediction, or CASP, said in a press release.
What's happening: At Monday's meeting of CASP, DeepMind announced that AlphaFold 2 — its second contender in the assessment that has happened every two years since 1994 — can reliably and accurately predict protein structures to within the width of an atom.
- CASP teams are given the sequences of proteins or parts of proteins over the course of a few months and submit the predicted structures.
- About two-thirds of the time, AlphaFold accurately predicted the protein structure on par with X-ray crystallography and cryo-electron microscopy, tried-and-true experimental techniques for determining protein structures that are expensive, time-consuming and a scientific art form.
- AlphaFold is a deep-learning network trained on about 170,000 protein structures. One area where the system struggled is with groups of proteins that can distort each other's shape, Nature News reported.
Keep in mind: Determining a protein's structure is a big step, but just one in the process of developing new drugs.
- "But DeepMind’s methods could be a way of determining whether a clinical trial will fail because of toxic reactions or other problems, at least in some cases," NYT's Cade Metz writes.