Figure 1

Description of three different protocols for approximating the binding affinity of a pHLA complex. Example input structures are visualized in the first row. The second row (orange stripe) shows the feature extraction phase of the scoring where ref2015 energy terms are extracted (Supplementary Table S1). The score calculation and training phase are indicated in row 3 (green stripe). (a) For ref2015-score, standard ref2015 energies are calculated for the complex, receptor, and ligand. They are then used to derive the binding energy with the Eq. (1) (b) For standard-pHLA-score, standard features are extracted from the complex; scoring is done using trained random forest regression models. (c) For the 3pHLA-score, per-peptide-position features are extracted from the structure of the complex; scoring is done using trained random forest regression models.