Evaluating Statistical Multiple Sequence Alignment in Comparison to Other Alignment Methods on Protein Data Sets.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multiple sequence alignment (MSA) is crucial for protein structure prediction, family identification, and phylogeny.
• Statistical coestimation of alignments and trees is theoretically rigorous but lacks validation on biological data.

Purpose of the Study:

• To evaluate the accuracy of popular protein alignment methods, including BAli-Phy, on biological and simulated datasets.
• To assess the performance of statistical coestimation in real-world bioinformatics applications.

Main Methods:

• Comparative analysis of BAli-Phy against other popular protein alignment tools.
• Extensive testing on 1192 biological and 120 simulated protein sequence datasets.
• Significant computational resources (over 230 CPU years) dedicated to BAli-Phy analyses.

Main Results:

• BAli-Phy demonstrated superior precision and recall on simulated datasets compared to other methods.
• BAli-Phy exhibited consistently lower recall on biological benchmarks than many alternative methods.
• A notable underalignment tendency of BAli-Phy on biological data, absent in simulated data.

Conclusions:

• BAli-Phy's performance varies significantly between simulated and biological data, suggesting potential issues like model misspecification or reference alignment errors.
• The findings highlight the need for further investigation into the causes of BAli-Phy's underalignment on biological sequences.
• Understanding these discrepancies is vital for improving the reliability of evolutionary and structural alignments in bioinformatics.