Hidden Markov models-based system (HMMSPECTR) for detecting structural homologies on the basis of sequential...

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Identifying structural homologs is crucial for understanding protein function and evolution.
• Existing methods face challenges in accurately predicting protein structures, particularly for complex targets.

Purpose of the Study:

• To introduce HMMSPECTR, a novel tool for detecting putative structural protein homologs.
• To evaluate the performance of HMMSPECTR against challenging protein targets from the CASP4 competition.

Main Methods:

• HMMSPECTR utilizes a data warehouse of hidden Markov models (HMMs) and alignment libraries.
• It incorporates a search program, secondary structure prediction, and a protein selection module.
• The tool was tested using protein targets from the CASP4 competition and publicly available Protein Data Bank data.

Main Results:

• HMMSPECTR 1.02 and 1.02ss versions demonstrated superior performance compared to the best CASP4 results in 64% and 79% of cases, respectively.
• Performance improvements were most pronounced for complex targets (complexity 4), indicating enhanced fold recognition capabilities.
• The tool achieved better results based on root-mean-square deviation (r.m.s.d.) and/or alignment length.

Conclusions:

• HMMSPECTR is an effective tool for identifying structural protein homologs.
• The automated versions of HMMSPECTR offer significant improvements in accuracy and efficiency for protein structure prediction.
• The tool shows particular promise for tackling difficult fold recognition problems in structural biology.