Figure 4. Expanding annotation coverage in the human proteome.

(A) Comparison of PARSE predictions for AlphaFold structures in the human proteome to EC number annotations in Uniprot, where available. Proteins are labeled as EC mismatch if the prediction does not match known annotations at any EC level, and putative new EC annotations are proteins with no EC numbers assigned in Uniprot. For these new hypotheses, we show the top 10 predicted enzyme classes at the third EC level. (B) Likely inactive PI-PLC with mutant catalytic residue H356T correctly not annotated by PARSE, and (C) putative class-C beta-lactamase predicted by PARSE. For both examples, reference structures from PDB are shown in green and query structures predicted by AlphaFold are shown in cyan. Residues identified as functional by PARSE are shown as sticks, and residues aligned to CSA residues but not annotated by PARSE are shown as lines. Catalytic residues are labeled using PDB numbering, and mismatches between query and reference are highlighted in orange. Proteins are aligned and RMSD is computed using catalytic residues only, including both backbone and side chain atoms.