Assessing strategies for improved superfamily recognition

Area of Science:

• Genomics
• Structural Biology
• Bioinformatics

Background:

• Public repositories contain millions of genome sequences and thousands of protein structures.
• Mapping structural data to genomic sequences is crucial for understanding protein function.
• Sequence-based methods are increasingly powerful for inferring structure from sequence.

Purpose of the Study:

• To review and assess sequence-based strategies for structural annotation of genomes.
• To describe the protocol for providing CATH structural annotations for completed genomes.
• To evaluate Hidden Markov Model (HMM) technologies for protein superfamily recognition.

Main Methods:

• Utilized Hidden Markov Model (HMM) technologies for superfamily recognition.
• Developed and tested SAMOSA (sequence augmented models of structure alignments) models.
• Employed a CATH-ISL (CATH-Integrated Structural Library) expanded HMM library.
• Annotated protein sequences from 120 genomes across three kingdoms.

Main Results:

• A single-seed HMM library recognized 76% of remote homologs.
• SAMOSA models showed minimal gain in homolog recognition but improved alignment quality for very remote homologs.
• An expanded 1D-HMM library (CATH-ISL) increased coverage to 86%.
• Up to 70% of genes in 120 annotated genomes were assigned to CATH superfamilies.
• Recruitment of sequences expanded the CATH database eightfold.

Conclusions:

• Sequence-based methods, particularly HMMs and CATH-ISL, are effective for structural annotation of genomes.
• These methods enable large-scale assignment of genes to protein structural superfamilies.
• The approach significantly expands protein domain databases and aids in understanding genomic content.