Fig. 5: Analysis of potential factors influencing AF2 peptide docking performance.

a Peptide length does not affect docking performance: successful models are generated for long as well as short peptides (See Supplementary Fig. 8A for the distribution of peptide lengths in the LNR dataset). Values over 20 Å are indicated by triangles. b Alpha helical peptides are modeled particularly well. Cumulative plot of performance according to the secondary structure of the peptide. (See Supplementary Fig. 8b for the corresponding distribution of secondary structures). Helical, beta strand, and coiled peptides are colored in pink, blue and gray, respectively. c Examples of potential memorization by AF2 from similar monomer structures: Left: Highly accurate prediction where a precise coverage of the structure of the peptide-receptor interface is available (native PDB ID 1ssc⁷⁰ and memorized PDB ID 1a5p⁷¹). Center: Inaccurate template but successful modeling (native PDB ID 3ayu⁷² and memorized PDB ID 5ue4⁷³). Right: Inaccurate template and failed model (native PDB ID 2x72⁷⁴ and memorized PDB ID 5dgy⁷⁵). In the latter two cases the model is most probably not built based on the inaccurate template. For all panels, the native receptor is shown in white, and native peptide in black, the memorized peptide in blue, and the modeled peptide in green. d Cumulative plot of performance on a set of peptide-receptor complex structures that were excluded due to post-translational modifications or bound ligands that influence the peptide structure: AF2 modeling is not dramatically impaired. The LNR set is shown in yellow, and the PTM + LIG set is shown in black. Source data are provided as a Source Data file.