High-quality sequence clustering guided by network topology and multiple alignment likelihood.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Proteins are classified into homologous families from common ancestors.
• Phylogenetic analysis of homologous sequences reveals gene function and evolution.
• Distinguishing divergent homologs from partial homology due to domain rearrangements is challenging for current clustering methods.
• Existing methods require a priori parameter settings, which are not universally optimal for all gene families.

Purpose of the Study:

• To develop a robust strategy for clustering homologous protein sequences that accounts for evolutionary divergence and domain rearrangements.
• To create a method that does not require a priori parameter settings and is adaptable to various gene families.

Main Methods:

• A multi-stage clustering approach is employed, starting with sensitive pairwise similarity comparisons to form pre-families.
• Pre-families are refined into homogeneous clusters based on similarity network topology.
• Clusters are merged into families using multiple sequence alignments and model selection to optimize family definition and alignment quality.

Main Results:

• The proposed method, HiFiX, effectively clusters sequences homologous across their entire length.
• HiFiX demonstrates robustness against both sequence divergence and protein domain rearrangements.
• The method is computationally efficient, suitable for large-scale genomic datasets.

Conclusions:

• HiFiX provides a superior solution for protein sequence clustering, overcoming limitations of existing approaches.
• The method's ability to handle complex evolutionary scenarios makes it valuable for evolutionary and functional genomics studies.