Amino acid mutations can cause disease by affecting the interactions of proteins with other molecules, including other proteins, nucleic acids, and ligands. We previously demonstrated an improvement in phenotype prediction by inclusion of the structure of biological assemblies rather than just the monomers in a machine-learning model. However, features derived from assemblies depend on the accuracy of annotations of these assemblies in the Protein Data Bank (PDB). It is estimated that only 80-90% of assembly annotations for X-ray crystallographic structures in the PDB are correct. The observation of similar interfaces or assemblies in multiple crystal forms of a protein or members of a protein family has long been used by crystallographers to identify biologically relevant interactions within crystals. We previously benchmarked the observation of similar interfaces across multiple crystal forms of non-identical proteins associated with biological relevance and developed the Protein Common Interface Database (ProtCID) for providing clusters of similar interfaces of full-length homologous protein chains across multiple crystal forms.<br/><br/>We have extended this idea to clustering homodimeric and heterodimeric interfaces at the domain level, as defined by Pfam domains assigned to the whole PDB. Clustering at the domain level revealed many more clusters as well as larger clusters of similar interfaces than clustering at the full-length chain level. For example, a homodimer of the regulatory ACT domain is found in 35 crystal forms of 28 different enzymes in 39 PDB entries. For full-length proteins, the largest cluster contains only 6 crystal forms of 6 proteins in 9 entries. Analysis at the domain level has enabled new features in ProtCID: clustering of interactions of protein domains with peptides, nucleic acids, and small-molecule ligands. A crucial feature of ProtCID is the ability to download single archive files that contain coordinates for each interaction in a cluster and scripts for visualizing the structures in PyMol. We have added access to ProtCID data at the level of Pfam clans, and enabled a new search feature in ProtCID with which a user can identify possible domain-domain and domain/peptide interactions that are possible amongst a set of proteins. Finally, we have clustered oligomeric assemblies of homologous proteins, and will present some new data on the utility of this approach for identifying oligomeric biological assemblies in the PDB.

ProtCID and ProtCAD: PDB-wide clustering of protein-protein interfaces and protein assemblies for the annotation of biological assemblies

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.