Zhanglab forum

Hi, hello!
I sincerely want to ask, in the ResPRE training process, how to deal with the problem of extremely imbalanced positive and negative samples?

In all training sets, only a few residue pairs are in contact. Most of the residues are not touched. In this case, the training model will easily overfit. How to deal with this situation?

We did see the existence of imbalanced positive and negative samples. But we do not think that it is a problem.
Of course, you can set larger weights or other regular settings to try to make it more balance during the training.