HPmod (Human Proteome Model) database is a repository for human protein structure and function that automatically predicted by the state-of-the-art algorithm from Yang Zhang Lab. Protein structure models are predicted by D-I-TASSER, and the protein functions, including Gene Ontology (GO), Enzyme Commission (EC), and ligand-binding sites, are predicted by COFACTOR. The database contains 19,512 proteins collected from Uniprot, which can be classified as 12,236 single-domain proteins and 7,276 multi-domain proteins. (>>more about the HPmod database)
The current database contains 19,512 entries.
May 1, 2023:
19,512 models for proteins of human proteome are generated by full D-I-TASSER modeling. The spatial restraints prediction
is created by AlphaFold2, AttentionPotential and DeepPotential with MSA created by DeepMSA2, and initial conformations are from LOMETS3 templates, then a Replica-Exchange
Montre Carlo (REMC) simulation is used for constructing 5 models. Multi-domain protein is handling by a new developed domain dealing module,
where the full-length sequence is first spilt by FUpred and ThreaDom to domain-level
sequences, individual domain-level spatial restraints are predicted by AlphaFold2, AttentionPotential and DeepPotential again, then
combined to full-length-level spatial restraints. The full-length D-I-TASSER model is created by REMC simulation guided by the predicted spatial restraints.
In addition, 34,968 domain-level models are also created by D-I-TASSER. The function prediction for each full-length and domain-level D-I-TASSER model is adopted by COFACTOR.