For now, at least we have "grep" and "diff".

Virus are like 30k base pairs. It's a quite short sequence for neural networks to tackle.

You can already use GPT-2 to generate compiled code already. You train it on known viruses and have it predict new variants.

You can also for better sample efficiency, jointly train to predict for any measurable characteristics you like, for example mortality or propagation speed, you then synthesize the various proposed DNA sequences, and inject them into hosts. Compare the results with the prediction and update your models accordingly.

Of course you will have plenty of non functioning organism, but that's probably a more efficient way of exploring this search space than what nature does with random mutations and cross-overs.

It's like giving virus the ability to virtually meet-up in the hall of fame to evolve faster.

will that disassembly tool include regulator genes and all the protein-protein interactions that influence said regulator genes, ad infinitum? there's so much more complexity to life than gene sequences themselves.