Related Experiment Videos
Ancestral sequence reconstruction in primate mitochondrial DNA: compositional bias and effect on functional inference
Neeraja M Krishnan1, Hervé Seligmann, Caro-Beth Stewart
1Biological Computation and Visualization Center, Department of Biological Sciences, Louisiana State University, USA.
Molecular Biology and Evolution
|July 2, 2004
Summary
Reconstructing ancestral DNA sequences using optimization methods like ML and parsimony can introduce bias. Bayesian ensemble methods offer more accurate and less biased ancestral nucleotide frequency reconstructions, improving functional predictions.
Area of Science:
- Evolutionary biology
- Molecular evolution
- Bioinformatics
Background:
- Reconstructing ancestral DNA sequences aids in understanding evolutionary events, molecular function changes, adaptation, and convergence.
- Maximum likelihood (ML) and parsimony are common methods, but both suggest directional nucleotide frequency changes in primate mitochondrial DNA (mtDNA).
Purpose of the Study:
- To investigate the differences between parsimony and ML in ancestral sequence reconstruction.
- To evaluate the accuracy and bias of various ancestral sequence reconstruction methods, including Bayesian approaches.
- To assess the impact of reconstruction bias on functional predictions of ancestral molecules.
Main Methods:
- Development of computationally simple "conditional pathway" methods with varying substitution allowances.
- Evaluation of the Bayesian posterior frequency distribution of reconstructed ancestral states.
- Analysis of primate mitochondrial cytochrome b (Cyt-b) and cytochrome oxidase subunit I (COI) genes.
- In silico evaluation of helix-forming propensities for conserved pairs in inferred ancestral primate mitochondrial tRNA sequences.
Main Results:
- ML reconstructions showed greater divergence from tip sequences than parsimony.
- Bayesian posterior ensemble frequency reconstructions more closely resembled extant nucleotide frequencies.
- Simulations suggested deterministic bias in optimization-based methods (parsimony, ML, Bayesian MAP) due to uncertainty.
- Averaging Bayesian credible ancestral sequences yielded less biased nucleotide frequencies.
- Simpler conditional pathway methods provided unbiased reconstructions with slightly reduced likelihood values.
- Bayesian ancestral tRNA reconstructions were more compatible with canonical base pairing than other methods.
Conclusions:
- Optimization-based ancestral sequence reconstruction methods can introduce significant nucleotide bias.
- Bayesian ensemble approaches provide more accurate and less biased ancestral nucleotide frequency reconstructions.
- Biased ancestral sequence reconstruction can lead to inaccuracies in functional predictions, particularly for molecules like tRNA.