Accuracy estimation and parameter advising for protein multiple sequence alignment.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multiple sequence alignment (MSA) is crucial for understanding protein and DNA evolution.
• Accurate MSA is essential for downstream analyses, but its quality is difficult to assess without a reference.
• Parameter selection for alignment scoring functions significantly impacts MSA accuracy.

Purpose of the Study:

• To develop a novel, general method for estimating MSA accuracy without a reference.
• To introduce a parameter advising task for selecting optimal alignment scoring function parameters.
• To improve the accuracy of computed MSAs through automated parameter optimization.

Main Methods:

• Developed a feature-based accuracy estimator (Facet) using polynomial functions of twelve alignment features.
• Employed mathematical optimization to minimize the error of the accuracy estimator against true accuracy.
• Introduced novel, fast-to-evaluate nonlocal feature functions for alignment quality assessment.
• Formulated new regression approaches for learning the accuracy estimator from data.
• Determined optimal parameter sets for alignment scoring functions.

Main Results:

• The Facet accuracy estimator significantly outperforms prior approaches for assessing alignment quality.
• The parameter advisor based on Facet achieved over 27% improvement in accuracy on challenging benchmarks compared to default parameters.
• The novel feature functions capture nonlocal alignment properties efficiently.

Conclusions:

• Facet provides a robust and general method for estimating MSA accuracy without a reference.
• The parameter advising approach effectively optimizes alignment scoring parameters, leading to substantial accuracy gains.
• This work advances the field of bioinformatics by improving MSA quality assessment and generation.