AI learns COVID’s language to predict its next move

MIT researchers have recognized a similarity between the way the immune system deciphers a virus and the way people comprehend one another’s thinking through language.

In a research report running in Science, the team details how principles behind these parallel modes of communication could be used with AI to predict which mutating strains of some viruses, including SARS-CoV-2, will become the most harmful.

Such predictions, the researchers suggest, could help guide refinements of existing vaccines. The idea would be to head off serious gathering threats without getting thrown off the scent by strains unlikely to proliferate.

The process by which viruses mutate into strains the immune system can’t recognize is called viral escape.

In the paper, biological engineer Bryan Bryson, PhD, computer scientist Bonnie Berger, PhD, and co-authors describe their work modeling viral escape with AI algorithms originally developed for natural language processing.

The NLP system was designed to process both syntax (i.e., sentence or clause construction) and semantics (i.e., meaning).

For the viral escape prediction project, the researchers trained their experimental system on several thousand genetic sequences of spike proteins, artistic depictions of which are now ubiquitous. To get the model to differentiate COVID-19 from cold or flu viruses, the training sets included spikes from coronaviruses other than SARS-CoV-2.  

“Semantic landscapes for these viruses predicted viral escape mutations that produce sequences that are syntactically and/or grammatically correct but effectively different in semantics and thus able to evade the immune system,” Bryson and co-authors explain. “We identified escape mutations as those that preserve viral infectivity but cause a virus to look different to the immune system, akin to word changes that preserve a sentence’s grammaticality but change its meaning.”  

In conversational coverage of the work by IEEE Spectrum, co-author Brian Hie, a PhD candidate at MIT, says he hopes the model will eventually be able to predict viral mutations before they arise.

“That’s a moonshot kind of goal for this line of research: vaccinating against future forms of the virus,” Hie says.

The study is available in full for free.