Measuring the fit of sequence data to phylogenetic model: allowing for missing data

Area of Science:

• Evolutionary biology
• Bioinformatics
• Computational phylogenetics

Background:

• Assessing data-model fit is crucial in phylogenetic and evolutionary studies.
• Standard statistical tests (e.g., G, X²) require complete sequence data, excluding sites with missing information (gaps, ambiguous residues).
• Removing incomplete data sites is often impractical and leads to significant data loss.

Purpose of the Study:

• To develop a method for estimating site-pattern probabilities directly from alignments containing missing data.
• To enable accurate model fit assessment in phylogenetics without discarding incomplete data.
• To extend the applicability of statistical tests to real-world molecular sequence alignments.

Main Methods:

• Utilizing iterative Maximum Likelihood (ML) estimators to compute site-pattern probabilities for columns with missing data.
• Employing Expectation-Maximization (EM) or Newton algorithms for optimization under standard independent and identically distributed (i.i.d.) assumptions.
• Comparing estimated probabilities with model expectations using likelihood-ratio (G) statistics or modified chi-squared (X²) tests.

Main Results:

• Demonstrated that ML estimators can directly handle missing data in sequence alignments.
• Developed a statistically sound approach to calculate G and X² statistics on alignments with missing data.
• Showcased the method's applicability to codon and paired-site models, including those with site-rate variability and Hadamard conjugations.

Conclusions:

• The proposed ML-based method effectively addresses the challenge of missing data in phylogenetic sequence alignments.
• This technique allows for more robust and comprehensive model fit assessment, preserving data integrity.
• The method enhances the utility of phylogenetic and evolutionary analyses by accommodating real-world data complexities.