Related Experiment Videos
Heterotachy and long-branch attraction in phylogenetics
Hervé Philippe1, Yan Zhou, Henner Brinkmann
1Canadian Institute for Advanced Research, Centre Robert-Cedergren, Département de Biochimie, Université de Montréal, Succursale Centre-Ville, Montréal, Québec H3C3J7, Canada. herve.philippe@umontreal.ca
BMC Evolutionary Biology
|October 8, 2005
Summary
Maximum Likelihood (ML) outperforms Maximum Parsity (MP) in phylogenetic tree reconstruction, especially when considering heterotachy and rate variation across lineages. ML is more accurate for real data analysis, even with complex evolutionary rate changes.
Area of Science:
- Phylogenetics
- Computational Biology
- Evolutionary Biology
Background:
- Maximum Parsimony (MP) was historically preferred for phylogenetic tree inference but is sensitive to Long Branch Attraction (LBA).
- Recent studies suggested MP's resilience to heterotachy (shifts in evolutionary rates), challenging Maximum Likelihood (ML).
- Previous simulations often used unrealistic assumptions, like equal terminal branch lengths, limiting applicability to real-world evolutionary scenarios.
Purpose of the Study:
- To evaluate the performance of MP and ML phylogenetic methods under more realistic simulation conditions.
- To investigate the impact of both heterotachy and rate variation across lineages on tree reconstruction accuracy.
- To clarify the relative strengths of MP and ML when faced with complex evolutionary rate heterogeneity.
Main Methods:
- Performed simulations incorporating both within-site rate variation (heterotachy) and rate variation across lineages.
- Generated heterotachous datasets using a protocol adapted from Kolaczkowski and Thornton.
- Assessed the accuracy of MP and ML tree reconstructions under varying levels of heterotachy and lineage-specific rates.
Main Results:
- Heterotachy decreases ML accuracy, violating model assumptions, irrespective of lineage rate variation.
- MP's accuracy can increase or decrease with heterotachy, depending on relative branch lengths, indicating sensitivity.
- ML outperforms MP in Long Branch Attraction (LBA) scenarios, except under extreme heterotachy levels.
Conclusions:
- Maximum Likelihood (ML) is consistently more accurate than Maximum Parsity (MP) for realistic combinations of heterotachy and lineage-specific rate variation.
- ML is recommended for analyzing real phylogenetic data due to its superior accuracy and ability to handle various biological heterogeneities.
- Mixture models within a probabilistic framework offer a way to mitigate the confounding effects of heterotachy on tree reconstruction.