Related Experiment Videos
Very fast algorithms for evaluating the stability of ML and Bayesian phylogenetic trees from sequence data
Peter J Waddell1, Hirohisa Kishino, Rissa Ota
1Department of Statistics and Department of Biological Sciences, University of South Carolina, Columbia, SC 29208, USA. waddell@stat.sc.edu
Genome Informatics. International Conference on Genome Informatics
|October 23, 2003
Summary
The RELL method offers a fast and accurate approximation for assessing evolutionary tree uncertainty, outperforming some bootstrap methods. This approach improves Bayesian posterior probability estimates, especially when models break down, making tree inference more reliable.
Area of Science:
- Phylogenetics
- Computational Biology
- Evolutionary Biology
Background:
- Evolutionary trees are crucial for sequence analysis, homology searches, and structural modeling.
- Assessing stochastic error in tree estimation is vital, but computationally intensive methods like bootstrapping are often prohibitive.
- Existing methods struggle with accuracy and speed for large datasets or complex evolutionary models.
Purpose of the Study:
- To evaluate the performance of the resampling of estimated log likelihoods (RELL) method for evolutionary tree uncertainty assessment.
- To improve the accuracy of Bayesian posterior probability (BPP) estimates by refining the BIC approximation.
- To investigate the impact of model breakdown on tree inference and propose solutions using resampling techniques.
Main Methods:
- Comparison of the RELL method against bootstrap (BP) methods for tree proportion approximation.
- Enhancement of the Bayesian Information Criterion (BIC) approximation for BPP using information matrix determinants.
- Analysis of mammalian mitochondrial amino acid sequences to assess model breakdown effects.
- Exploration of incorporating bootstrap and other resampling methods with Markov Chain Monte Carlo (MCMC) procedures.
Main Results:
- RELL approximates bootstrap proportions effectively and is computationally faster than traditional bootstrapping.
- The refined BIC approximation yields BPP estimates closely matching MCMC results.
- Model breakdown in deep evolutionary time leads to overly peaked BPP values, necessitating alternative approaches.
- BP and BPP estimates can remain reliable indicators of the species tree even with site-specific evolutionary processes.
Conclusions:
- The RELL method provides a computationally efficient and accurate alternative for assessing evolutionary tree uncertainty.
- Improved BPP estimation methods enhance the reliability of phylogenetic inference, particularly under model misspecification.
- Integrating resampling techniques with MCMC offers robust solutions for phylogenetic analysis in the face of evolutionary complexities.