One of the leading problems is that we don't know how to pick the right neoantigens (mutant cancer proteins) which can trigger the optimal immune response.

Moderna used a simple heuristic, "how likely is this antigen to show up on the cell surface?" which makes sense, since an antigen has to show up on the cell surface for the immune system to see it. But there's so much missing from this model, and we just don't have enough data. Failures in clinical trials of mRNA vaccines may well be caused by this problem. (Yes, this is a good problem for AI, if we can get enough data).