Estimation of relative effectiveness of phylogenetic programs by machine learning

Area of Science:

• Bioinformatics
• Computational Biology
• Evolutionary Biology

Background:

• Phylogenetic tree reconstruction quality varies based on input sequence alignment.
• Evaluating reconstruction quality often requires comparison with known organismal phylogenies.

Purpose of the Study:

• To develop a predictive model for the quality of protein phylogeny reconstruction from sequence alignment features.
• To assess the feasibility of predicting the optimal phylogenetic reconstruction method (Fitch-Margoliash vs. UPGMA) for a given alignment.

Main Methods:

• Utilized the Fitch-Margoliash (FM) method for phylogeny reconstruction.
• Employed a random forest classifier to predict reconstruction quality based on alignment features.
• Trained and tested the predictor using alignments of orthologous series (OS) with known phylogenetic outcomes.

Main Results:

• Achieved over 80% precision in predicting the quality of phylogeny reconstruction.
• Demonstrated a principal possibility of predicting the better method (FM or UPGMA) for specific alignments.
• The predictor identified UPGMA as superior for 56% of alignments where it truly performed better, compared to 34% in the test set.

Conclusions:

• Alignment features can reliably predict protein phylogeny reconstruction quality.
• Predicting the optimal phylogenetic reconstruction method based on alignment characteristics is feasible.
• This approach can improve the accuracy and efficiency of phylogenetic analyses in bioinformatics.