Fig. 6
Relationship between median pLDDT scores and protein sequence length of AI-predicted models. Median pLDDT scores are plotted against the sequence length of (A) AlphaFold 2, (B) OpenFold, and (C) ESMFold models. Note for panel C: The median pLDDT values for proteins larger than 400 amino acids exhibit a more uniform distribution. This is because models for these proteins were generated using BioNeMo, which provides pLDDT values as percentages. In contrast, pLDDT values obtained from the Meta API, which was used for proteins up to 400 amino acids, are expressed as relative numbers. Since PDB files allow for a limited five-digit space to store pLDDT values, relative numbers (e.g., 0.953) contain one less digit than percentages (e.g., 95.32).
