predict_hyp could support unknown amino acids symbols X (any), B (D or N) and Z (E or Q) if they occur up to once or even twice in subsequence.
This could proceed by generating all possible subsequences (this is why a limit of X is needed 20 20 is manageable but 20 20 * 20 is too much per subsequence), predicting Hyp probability in them and retuning the probability that is the min, max or floor(median) of predictions based on an additional argument in the function.
predict_hyp
could support unknown amino acids symbols X (any), B (D or N) and Z (E or Q) if they occur up to once or even twice in subsequence.This could proceed by generating all possible subsequences (this is why a limit of X is needed 20 20 is manageable but 20 20 * 20 is too much per subsequence), predicting Hyp probability in them and retuning the probability that is the min, max or floor(median) of predictions based on an additional argument in the function.