Fig. 2. Design and benchmarking study of the 2DP fingerprint. (a) Equation for the calculation of 2DP entries for each atom category. (b) Performance of 2DP nearest neighbor searches for recovering various peptide types. Frequency histogram of AUROC values for recovering up to 100 3DP-nearest neighbours of each of the 2073 peptides up to 50 residues in the Protein Data Bank from all other entries within ±10% size of each query (blue curve), recovering 309 AMP sequences from 50 AMP sequences among 8815 decoys sequences in the UniProt databank (all limited to 50 residues, red curve), and recovering the closest BLAST analogs (with bit score > 19, 110-285 BLAST analogs) of 50 different 13-mer peptides from 8000 randomly scrambled peptide sequences of the query peptide (green curve).
