Accurate domain identification with structure-anchored hidden Markov models, saHMMs

Area of Science:

• Genomics and Bioinformatics
• Structural Biology
• Computational Biology

Background:

• DNA sequencing speed outpaces functional and structural annotation of gene products.
• Accurate protein sequence annotation is vital for understanding function via similarity comparisons, which is challenging at low sequence identities.

Purpose of the Study:

• To enhance protein domain identification accuracy by leveraging conserved 3D structures over sequences.
• To develop a reliable tool for assigning protein domains to their correct families, even for distantly related sequences.

Main Methods:

• Constructed 850 structure-anchored hidden Markov models (saHMMs) from structure-anchored multiple sequence alignments of homologous domains.
• Utilized the Structural Classification of Proteins (SCOP) database for domain family definitions.

Main Results:

• Achieved 99.0% accuracy in domain identification using saHMMs against SCOP database.
• saHMMs demonstrated higher coverage (11%) than Pfam_ls HMMs and fewer errors per query than BLAST and PSI-BLAST.
• Successfully annotated 530 unknown domains in human proteins across 102 families.

Conclusions:

• The saHMM database is a versatile and reliable tool for protein domain identification and homology assignment.
• saHMMs facilitate the annotation of distantly related sequences and link findings to high-quality crystal structures.
• The saHMM database is accessible via the FISH server.