Video: we can get critical information on virology from new methods
In some ways, virology and pathology have implications for how we are going to benefit from the fruits of AI.
For example, the accumulated data can be useful in any number of ways. Adding to what we have already seen on medical applications, there’s more from David Gifford, an MIT professor and CEO of a company called Think Therapeutics, which pioneers some of the types of research he’s talking about, including both genetics and immunology.
Or rather, the intersection of those two disciplines.
Gifford starts out touting the value of vaccines, which isn’t difficult in the post-Covid age.
“What drug has saved the most lives throughout history?” he asks. “Now, you might think, oh, penicillin (or) an antibiotic, right? That’s obvious… not so: vaccines as a drug class have saved more lives throughout history than any other drug class – a billion lives and counting.”
As for Covid-19, the king of the viral menaces, he notes that many of us “still walk around a little bit in terror,” in introducing ways to combat viral loads in general.
You could say that, as a rule, antiviral work is the process of enabling antibodies. The problem, as Gifford puts it, is that viruses can ‘escape’ from the roving healers that consume infected cells.
“They run away from current therapeutics, because the way vaccines work at present, in licensed vaccines, is to induce an antibody response against the face of the virus, and the face changes over time. And we actually don’t know exactly where it’s going to go in the future, nor is it really easy to generate a single antibody that protects against all possible future variants. So what’s the alternative? And how can we get around this fundamental challenge? Well, here’s your proposal: we’re not going to go after the face of the vaccine, we’re going to go after its heart. We’re going to go after the parts of the virus that can’t change (what makes it work).”
Referring to “killer T-cells,” he shows us a video of one attacking some other cell by stabbing wildly at its surface, while also introducing models for working with an epitope, the part of an antigen molecule that an antibody attaches to, and alleles, variations in the sequencing of nucleotides in a long DNA molecule.
Gifford also suggests that n-times coverage can have a statistical effect.
“Our programming is successful,” he says. “We’re getting the killer T-cells to recognize bits of the virus in the mouse.”
On the CSAIL blog, Rachel Gordon provides more detail on the “OptiVax” system that Gifford covers in his talk. Gordon calls it “a combinatorial machine learning system that selects peptides (short strings of amino acids) that are predicted to provide high population coverage for a vaccine,” writing:
“The design system… introduces methods for designing new peptide vaccines, evaluating existing vaccines, and augmenting existing vaccine designs. In this system, peptides are scored through machine learning by their ability to be displayed to elicit an immune response, and are then selected to maximize population coverage of who could benefit from the vaccine.”
“You can see its superior population coverage,” Gifford says, in demonstrating the method. “We’re going to immunize on day zero and day 21, three weeks later, and then we’re going to expose this mouse that’s been immunized to a beta variant of COVID, which is very pathogenic and has a host range that includes mice. So when we do this, we see … in the top plot here, that we get a good immune response. Our programming is successful.”
Closing out the presentation, Gifford brings the focus to a broader use of AI in this field:
“AI provides novel medical solutions,” he says. “It’s just beginning, not only for vaccines, but as we also heard in the previous talk, for other kinds of therapeutic modalities … it’s a very exciting time.”
Read the full article here
