Performance of predictive models using different sets of residue interactions.

The predictive models were optimized using the combined set of interactions and evaluated with nested cross-validation as described in the Methods. The residue interactions used to describe the coiled-coil interaction were varied as descried in the text. We also applied recursive feature elimination to find a smaller subset of key features that retained good prediction performance. The reduced number of features, and the model performance with this number of features, are given in parentheses.Performance of predictive models using different sets of residue interactions.