News for hackers. Stuff that matters.
Re:Pfizer vaccine appears effective against mutation in new ... by sudosysgen on Saturday January 09, 2021 @01:47PM (#25703212)
You can make multiple spike proteins that target the same receptor.

This generally happens when a virus from another animal (and thus with a slightly different version of the receptor) adapts to humans.

Basically, a gradient descent from a very different starting point ending up in a different local minima.

The question is, is the spike protein at a local minima (very probably quite close to it), and could it jump to another local minima then? Most likely not.

It's possible it will evolve to make our antibodies slightly less effective, likely as a trade off for infectiousness, but incredibly unlikely the spike protein would evolve to be unrecognizable to our immune systems. Historically, this doesn't happen all that often, except for viruses with chronic infections over years that can do a deeper search over the gradient, but even then it generally takes years to decades.

Re:Pfizer vaccine appears effective against mutation in new ... by lawrenceyan on Saturday January 09, 2021 @03:55PM (#25704638)
Given the non-linear dynamics that proteins must live under, I like the usage of local minima as an analogy. It feels apt.

Generally though, within the context of machine learning, one of the benefits of gradient descent, especially when stochastic, is that we can get past those local minima humps. Does this hold less true with respect to the process of sequence mutation that viruses go through?

Re:Pfizer vaccine appears effective against mutation in new ... by jhrmnn on Sunday January 10, 2021 @03:50AM (#25712482)
Another way how to characterize the "context of [deep] machine learning" is that it is the regime of high temperature (stochasticity) as a result of small batches and/or large learning rates, and/or the regime of high momentum (Adam et al.). The evolutionary dynamics in the genetic space is (intuitively) at much lower temperature (shorter jumps) and with no momentum (because of the evolutionary pressure).
Re:Pfizer vaccine appears effective against mutation in new ... by sudosysgen on Saturday January 09, 2021 @08:55PM (#25707778)
The difference in viruses is that if you get on the hump, you are outcompeted by other viruses that stayed in the minimum, which makes it a lot more difficult.
Re:Pfizer vaccine appears effective against mutation in new ... by lawrenceyan on Sunday January 10, 2021 @03:59PM (#25719735)
That makes a lot of sense. Thanks!
Re:Pfizer vaccine appears effective against mutation in new ... by dimgl on Saturday January 09, 2021 @03:50PM (#25704605)
This is extremely interesting. Thanks for this context. Detailed and thoughtful explanations like this help to understand the news.