Ask HN: Could reinforcement learning be used to create vaccines?

[fishbone]

My understanding is that reinforcement learning can be used to train neural networks to become chess masters.

Would it be possible to pit a virus generator NN vs a vaccine generator NN to create vaccines? Seems like you could train a NN that could generate a vaccines to combat viruses that we haven’t even seen yet.

[burning_hamster]

Anti-viral vaccines are not anti-viral drugs. The main problem is not figuring out an exploit in the virus biology that could be leveraged by a pharmacological agent to disrupt the viral life cycle. The main problem is

1. finding a part or form of the virus that elicits a strong immune response, which

2. does not cause illness on its own.

Optimization criteria 1. and 2. are in direct conflict with each other as our immune systems are pretty efficient at ignoring stuff that is foreign but not harmful (otherwise you would develop an immune reaction to all kinds of things which we would then call an allergic reaction). So you need to find the right balance, and what that balance is depends on the human response.

[burning_hamster]

Now that does not mean that one could not think about automated ways of finding vaccine candidates, but instead of modelling (implicitly or explicitly) viruses and vaccines, you would need to model virus-derived particles and human immune responses (plus the effect of adjuvants, etc.).

[fishbone]

Good information, so we would have a Neural Net that takes a vaccine input and predicts the human (or animal) immune response, and another neural net that generates the vaccine inputs. I suppose the problem would be lack of labeled training data, which is why I was thinking about the chess example which works by simply knowing the rules of the game.