Identifying Clusters of High Confidence Homologies in Multiple Sequence Alignments.

Area of Science:

• Bioinformatics
• Evolutionary Biology
• Computational Biology

Background:

• Multiple sequence alignment (MSA) accuracy is critical for downstream evolutionary inference.
• Errors and uncertainty in MSAs can lead to significant problems, including false positive selection signals and long branch attraction artifacts.
• Current methods, such as column filtering, have shown mixed success and can sometimes be detrimental to phylogenetic studies.

Purpose of the Study:

• To develop a novel method for addressing MSA uncertainty and error.
• To introduce Divvier, a software tool employing a graph-based clustering approach for MSA analysis.
• To improve the reliability of evolutionary inference by accurately identifying homologous characters.

Main Methods:

• Implemented a graph-based clustering method using a probabilistic model.
• Developed the Divvier software to identify clusters of characters with statistically supported homology.
• Offered options for partial filtering or representing clusters as new columns ('divvying').

Main Results:

• Divvier demonstrated superior performance on real and simulated data compared to existing filtering software.
• The software successfully retained more true pairwise homology calls and removed more false positive calls.
• Divvier mitigated long branch attraction artifacts and reduced tree estimation variation caused by MSA uncertainty.

Conclusions:

• Divvier offers a robust solution for handling MSA uncertainty and errors.
• The graph-based clustering approach significantly improves the accuracy of homology detection.
• This method enhances the reliability of phylogenetic analyses and evolutionary inference.