Cloud Hosting

A team of San Diego scientists is harnessing artificial intelligence to understand why COVID-19 symptoms can vary dramatically from one person to the next information that could prove useful in the continued fight against the coronavirus and future pandemics.

Researchers pored through publicly available data to see how other viruses alter which genes our cells turn on or off. Using that information, they found a set of genes activated across a wide range of infections, including the novel coronavirus. Those genes predicted whether someone would have a mild or a severe case of COVID-19, and whether they were likely to have a lengthy hospital stay.

A UC San Diego-led team joined by researchers at Scripps Research and the La Jolla Institute for Immunology published the findings June 11. The studys authors say their approach could help determine whether new treatments and vaccines are working.

When the whole world faced this pandemic, it took several months for people to scramble to understand the new virus, said Dr. Pradipta Ghosh, a UCSD cell biologist and one of the studys authors. I think we need more of this computational framework to guide us in panic states like this.

The project began in March 2020, when Ghosh teamed up with UCSD computer scientist Debashis Sahoo to better understand why the novel coronavirus was causing little to no symptoms in some people while wreaking havoc on others.

There was just one problem: The novel coronavirus was, well, novel, meaning there wasnt much data to learn from.

So Sahoo and Ghosh took a different tack. They went to public databases and downloaded 45,000 samples from a wide array of viral infections, including Ebola, Zika, influenza, HIV, and hepatitis C virus, among others.

Their hope was to find a shared response pattern to these viruses, and thats exactly what they saw: 166 genes that were consistently cranked up during infection. Among that list, 20 genes generally separated patients with mild symptoms from those who became severely ill.

The coronavirus was no exception. Sahoo and Ghosh say they identified this common viral response pattern well before testing it in samples from COVID-19 patients and infected cells, yet the results held up surprisingly well.

It seemed to work in every data set we used, Sahoo said. It was hard to believe.

They say their findings show that respiratory failure in COVID-19 patients is the result of overwhelming inflammation that damages the airways and, over time, makes immune cells less effective.

Stanfords Purvesh Khatri isnt surprised. His lab routinely uses computer algorithms and statistics to find patterns in large sets of immune response data. In 2015, Khatris group found that respiratory viruses trigger a common response. And in April, they reported that this shared response applied to a range of other viruses, too, including the novel coronavirus.

That makes sense, Khatri says, because researchers have long known there are certain genes the immune system turns on in response to virtually any viral infection.

Overall, the idea is pretty solid, said Khatri of the recent UCSD-led study. The genes are all (the) usual suspects.

Sahoo and Ghosh continue to test their findings in new coronavirus data as it becomes available. Theyre particularly interested in COVID-19 long-haulers. Ghosh says theyre already seeing that people with prolonged coronavirus symptoms have distinct gene activation patterns compared to those whove fully recovered. Think of it like a smoldering fire that wont die out.

The researchers ultimate hope isnt just to predict and understand severe disease, but to stop it. For example, they say, a doctor could give a patient a different therapy if a blood sample suggests theyre likely to get sicker with their current treatment. Ghosh adds that the gene pattern theyre seeing could help identify promising new treatments and vaccines against future pandemics based on which therapies prevent responses linked to severe disease.

In unknown, uncharted territory, this provides guard rails for us to start looking around, understand (the virus), find solutions, build better models and, finally, find therapeutics.

More:
Can artificial intelligence predict how sick you'll get from COVID-19? UC San Diego scientists think so - The San Diego Union-Tribune