Fig. 12

Summary of features adopted in our model. For each target protein with N residues, we have the original CCMpred map with the size of N × N. We calculate the ridge features for each point on the map to get 4 N × N matrices (2 N × N matrices after feature selection). In total, we have N × N × 5 (N × N × 3 after feature selection) 2D features. The secondary structure prediction from DeepCNF provides an N × 3 1D feature matrix. In addition, we have 2 map features (the sequence/residue ratio and CCMpred standard deviation) and 5 position features (1 residue index difference and 4 distances to protein ends). The data in this figure were generated from the protein 1AHQA