A new approach to vaccines with a machine learning twist could put an end to boosters and seasonal variant shots, according to MIT researchers. This “pan-variant” vaccine would ignore the virus itself but quickly control infections by cracking down on infected cells.
To be quite clear, this is still in animal testing and is nowhere near being deployed. But with COVID becoming a resident virus in the human population, longer-lasting solutions than occasional boosters for particularly bad strains are in demand.
The problem is that, as amazing as the mRNA vaccines are, they are reactive, not proactive: you see a variant, you sample its spike protein or some other distinctive feature and you slip it into the immune system so it knows to be on the lookout. It’s a bit like letting a rescue dog sniff the belongings of a lost hiker.
Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory wanted to find another, more enduring way to keep the body safe from COVID attack. A paper describing their findings was published today in the journal Frontiers in Immunology.
The team decided to punt on the idea of attacking the virus itself, because its most distinctive feature, the spike protein, is always changing. Instead, they focused on certain molecular signals that reliably appear on the surface of cells infected by the virus. If these could be spotted early and the immune system’s T cells deployed quickly, the infection would be stalled before it reached dangerous or potentially even infectious levels.
These surface signals, called human leukocyte antigens, and they present a variety of peptides to T cells, kind of like raising semaphore flags. If everything is in order, it’s the usual combination of familiar peptides and the T cell moves on. If something is wrong, a fragment of the virus may be hoisted up the flagpole, and the T cells open fire.
So what does machine learning have to do with all this? There’s a lot of data out there cataloguing the various proteins and amino acid chains found in COVID, and what they turn into once they’ve infiltrated a cell, and how the cells indicate using HLAs that they’re infected.
Machine learning algorithms are good at solving optimization problems like this, where lots of noisy data must be sorted through for specific combinations of qualities. In this case they designed algorithms to catalogue the relevant peptides and select about 30 that are present or “conserved” in all versions of the virus, but also are associated with HLAs, and are likely to be used as flags for T cells to see.
Transgenic mice given our versions of HLAs and this new vaccine showed a far more voluminous immune response in the short term after infection, and none died of the virus.
“This study offers proof in a living system, an actual mouse, that the vaccines we devised using machine learning can afford protection from the COVID virus,” said MIT PhD student Brandon Carter, one of the paper’s lead authors, in an MIT news article.
An interesting possible benefit is that immunocompromised people may get important protection from this approach while the mRNA vaccines don’t work for them. “Long COVID” sufferers too may get some relief in the form of a more comprehensive immune assault on their especially resilient infection.
As the study’s abstract puts it:
The undetectable specific antibody response in MIT-T-COVID-immunized mice demonstrates specific T cell responses alone can effectively attenuate the pathogenesis of SARS-CoV-2 infection. Our results suggest further study is merited for pan-variant T cell vaccines, including for individuals that cannot produce neutralizing antibodies or to help mitigate Long COVID.
It’s a promising line of inquiry and a great way to employ advances in computation in service of global health. But it’s also important to recognize it is still early days for the “pan-variant” option. For one thing, it may work with or against existing vaccines — what if the best peptides for the immune response vaccine are the ones targeted for destruction by mRNA priming? The two would be working at cross purposes. And too strong an immune response also runs the risk of collateral damage, mistaken culling of ambiguously signaling cells and the like.
But these are the good kind of questions — ones that are relevant because the basic function of the new vaccine appears to work. We’ll know more as the team progresses through more tests of this promising approach.