Multi-class prediction

[joaquimgomez]

Is there a plan to include multi-class prediction in some way? It is not something urgent because only one split in FLIP from the sub-cellular location has a multi-class target, but something to take account for the future.

[mheinzinger]

Definitely an interesting feature, thuogh, not super high priority, I guess.
At the end it could be as easy as: for each label-file that gets passed, one prediction head is added (with all prediction heads having one common trunk network that shares parameters).
For example, secondary structure would also be such a case as it has secondary structure in 3- and 8-states (and even "disorder" aka missing/unresolved residues; i.e., the mask that you are currently using also has some meaning, i.e., a residue was not experimentally resolved in the 3D structure due to noise. This can be predicted as it correlates to a certain extent with disorder).