Improved pairwise alignments of proteins in the Twilight Zone using local structure predictions

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Accurate protein homology detection is challenging in the 'twilight zone' of low sequence identity.
• Traditional global amino acid substitution matrices may not capture context-specific evolutionary pressures.
• Accurate pairwise sequence alignments are crucial for inferring orthologous relationships and protein modeling.

Purpose of the Study:

• To develop a novel model for amino acid substitution probabilities that incorporates structural context.
• To improve the accuracy of protein sequence alignments, particularly for distantly related proteins.

Main Methods:

• Introduced HMMSUM (HMMSTR-based substitution matrices), a set of 281 matrices based on predicted local protein structure.
• Utilized HMMSTR, a hidden Markov model, to predict local structure without requiring known protein structures.
• Implemented HMMSUM with local Dynamic Programming and the Bayesian Adaptive alignment method.

Main Results:

• HMMSUM-based alignments demonstrated favorable comparisons against alignments using BLOSUM, SDM, and HSDM matrices.
• Validation against remote homolog alignments from BAliBASE confirmed the efficacy of HMMSUM.
• The model effectively accounts for structural context in predicting amino acid substitution probabilities.

Conclusions:

• HMMSUM provides a more accurate approach to protein sequence alignment by integrating structural context.
• This method enhances the reliability of homology detection and pairwise sequence alignments in computational biology.
• HMMSUM offers a valuable tool for protein modelers and researchers studying evolutionary relationships.