When a new COVID-19 variant emerges, scientists race to determine its threat level. But often, answers come too late to influence public health decisions. Now, researchers at Harvard University have developed a faster, predictive approach using biophysics and artificial intelligence to stay ahead of viral evolution.
Forecasting Before the Threat Escalates
In two recent PNAS studies, a team from the Department of Chemistry and Chemical Biology introduced a multiscale model that predicts which viral variants are likely to become dominant. Led by Professor Eugene Shakhnovich, the researchers used the spike protein’s binding affinity and immune evasion properties—along with a key factor called epistasis—to forecast variant success before clinical data surfaces.
“Evolution isn’t linear,” said Shakhnovich. “Our model accounts for how mutations interact, allowing us to predict dominant strains ahead of time.”
Introducing VIRAL: AI-Powered Variant Detection
As reported by medicalxpress, building on this, the team developed VIRAL (Viral Identification via Rapid Active Learning), an AI-driven tool that prioritizes lab testing of the most dangerous spike protein mutations. VIRAL accelerates detection of high-risk SARS-CoV-2 variants by focusing experimental efforts where they matter most.
“At the start of a pandemic, we can’t test every mutation,” said co-author Dianzhuo Wang. “VIRAL helps us target those with the highest threat potential.”
Impact Beyond COVID-19
Simulations show VIRAL identifies dangerous variants five times faster than traditional methods—using less than 1% of the lab effort. The team envisions applying this framework to emerging viruses and even fast-evolving cancers.
“This isn’t just about COVID—it’s a proactive strategy for biological threats,” said co-author Marian Huot. Shakhnovich emphasized that federal research funding made this breakthrough possible and warned that budget cuts could jeopardize future innovations.