Related Experiment Videos
Evaluating the performance of a successive-approximations approach to parameter optimization in maximum-likelihood
Jack Sullivan1, Zaid Abdo, Paul Joyce
1Department of Biological Sciences, Initiative in Bioinformatics and Evolutionary Studies and Program in Bioinformatics and Computational Biology, University of Idaho, ID, USA. jacks@uidaho.edu
Molecular Biology and Evolution
|March 11, 2005
Summary
Successive approximations in maximum-likelihood (ML) phylogenetics are reliable. This faster method for estimating DNA phylogenies is safe and accurate, comparable to computationally intensive full optimizations.
Area of Science:
- Computational Biology
- Phylogenetics
- Molecular Evolution
Background:
- Phylogenetic tree estimation using maximum-likelihood (ML) often relies on approximations.
- These approximations involve estimating model parameters on an initial tree, then holding them constant during a tree search.
Purpose of the Study:
- To formally assess the effectiveness of the successive approximations method in ML phylogenetic inference.
- To compare the accuracy of successive approximations against full optimization methods.
Main Methods:
- Indirect evaluation: Replicate ML searches on real data with random initial parameter estimates.
- Direct assessment: Simulations comparing accuracy of successive approximations versus full optimization of all parameters.
Main Results:
- Replicate runs on real data consistently converged to the same topology and parameter estimates.
- Simulations showed no significant difference in accuracy between successive approximations and full optimization.
Conclusions:
- The successive approximations approach is reliable and not dependent on initial starting points for parameter estimation.
- This computationally efficient method is a safe and accurate alternative for ML estimation of phylogenetic topology.