STAT+: DeepMind and and a rival release dueling code for protein-folding AI

Computational biologists have been on tenterhooks for the past seven months, ever since DeepMind took a hammer to one of their field’s most persistent challenges: accurately predicting the 3D shape of a protein from its amino acid sequence.

The Alphabet-owned AI research outfit had developed a neural network that predicts protein structures with near-perfect accuracy, blowing the competition out of the water at a protein structure prediction contest called CASP. The field’s response was ebullient — and then, just as quickly, disgruntled. DeepMind didn’t share details of how its blockbuster method worked, limiting its disclosures to a press release and a 12-minute presentation at the contest.

Continue to STAT+ to read the full story…