Related Experiment Videos
Multiple maxima of likelihood in phylogenetic trees: an analytic approach
B Chor1, M D Hendy, B R Holland
1Institute of Fundamental Sciences, Massey University, Palmerston North, New Zealand.
Molecular Biology and Evolution
|October 6, 2000
Summary
Maximum likelihood (ML) tree selection can yield multiple optimal evolutionary trees, complicating phylogenetic analysis. This study identifies sequences causing these multiple optima, revealing limitations in current computational methods.
Area of Science:
- Phylogenetics
- Computational Biology
- Evolutionary Biology
Background:
- Maximum likelihood (ML) is a standard method for inferring evolutionary trees.
- The landscape of likelihood scores for phylogenetic trees, particularly the occurrence of multiple optima, remains poorly understood.
- Identifying conditions that lead to multiple optimal tree topologies is crucial for robust phylogenetic inference.
Purpose of the Study:
- To analytically investigate sequence data that generate multiple ML optima for phylogenetic trees.
- To develop a direct computation method for ML to identify such sequences.
- To assess the implications of multiple optima for phylogenetic analysis and computational algorithms.
Main Methods:
- Analytic study focusing on optimizing edge weights for a fixed tree topology.
- Direct computation of Maximum Likelihood (ML) values.
- Identification of sequence families exhibiting multiple ML optima.
Main Results:
- Discovery of large families of sequences that result in multiple ML optima, including continuous sets of optimal points.
- Demonstration that certain biological data sets can be best explained by multiple, significantly different phylogenies.
- Highlighting that current hill-climbing algorithms may fail to find the unique global ML optimum.
Conclusions:
- The likelihood surface for phylogenetic trees can possess multiple optima, challenging standard inference methods.
- Specific sequence characteristics can lead to data sets with multiple evolutionary interpretations.
- Existing computational approaches for finding the global ML tree may be insufficient due to the potential for multiple optima.