Algorithms are getting better and better at predicting protein structures from genomic data, including that of the Omicron variant of SARS-CoV-2, Wired reports.
It says that after the World Health Organization dubbed Omicron a variant of concern, researchers set about to try to figure out how its protein structure might differ from other SARS-CoV-2 variants. Researchers from the University of North Carolina at Charlotte used AI software to predict structural changes to the virus's spike protein's receptor binding domain, as they report in a preprint posted to BioRxiv. As Wired notes, their results track with what another group from the University of British Columbia found using cryo-EM structural analysis, presented in a separate preprint.
UBC's Sriram Subramaniam tells Wired that while direct measurements will remain the gold standard, AI predictions have and will continue to inform experimental work. "It's changed the way we think," Subramaniam adds there.
Wired cautions that algorithms are currently trained on databases that include structures that were the easiest to study, rather than diverse sampling and only provide a static view of the predicted stricture.