Improving model construction of profile HMMs for remote homology detection through structural alignment

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Bioinformatics

Background:

• Remote homology detection is a critical challenge in bioinformatics.
• Profile Hidden Markov Models (pHMMs) are a successful approach for homology detection.
• The impact of structural alignments on pHMM performance, particularly in low-identity protein regions (Twilight Zone), remains underexplored.

Purpose of the Study:

• To assess the impact of using structural alignments versus sequence alignments for training pHMMs.
• To evaluate performance differences in detecting remote homologies, especially in low-identity protein regions.

Main Methods:

• Utilized the SCOP database for experiments.
• Generated structural alignments using 3DCOFFEE and MAMMOTH-mult.
• Generated sequence alignments using CLUSTALW, TCOFFEE, MAFFT, and PROBCONS.
• Performed leave-one-family-out cross-validation and evaluated performance using ROC curves and t-tests.

Main Results:

• pHMMs derived from structural alignments significantly outperformed those from sequence alignments in low-identity regions (<20% identity).
• Structural alignments appear to better capture conserved evolutionary patterns, leading to higher quality pHMMs.
• Sensitivity remains a challenge for these low-identity regions, indicating room for improvement.

Conclusions:

• Structural alignments offer a significant advantage for training pHMMs in detecting remote homologies in low-identity protein regions.
• The findings suggest that focusing on structurally conserved motifs enhances pHMM accuracy.
• Further research is needed to improve the sensitivity of these methods for challenging low-identity regions.