Bingo. That would be really interesting. And useful.
There are probably already enzymes in this data set that have measurements of their behavior. Could this modelling approach be coaxed to find the one with the highest processivity? Or do we need more labeled data?