PhyloMissForest: a random forest framework to construct phylogenetic trees with missing data.

Area of Science:

• Evolutionary biology
• Bioinformatics
• Computational biology

Background:

• Phylogenetic trees are crucial for understanding evolutionary history and biodiversity.
• Missing data in phylogenetic analyses can lead to inaccurate evolutionary relationship inferences.
• Existing methods for handling missing data have limitations in applicability and accuracy.

Purpose of the Study:

• To introduce PhyloMissForest, a novel framework for imputing missing entries in phylogenetic distance matrices.
• To infer accurate evolutionary relationships despite the presence of missing biological data.
• To provide a robust and configurable solution for phylogenetic missing data imputation.

Main Methods:

• PhyloMissForest utilizes a random forest structure to infer missing data based on available information.
• The framework integrates multiple search strategies, machine learning, and phylogenetic techniques.
• A design of experiments methodology was employed for efficient hyperparameter tuning.

Main Results:

• PhyloMissForest generally outperforms state-of-the-art imputation techniques on real DNA data, especially with up to 60% missing data.
• Significant improvements in execution time were observed for amino acid data imputation.
• The framework demonstrates improved imputation accuracy on simulated data with large percentages of missing values.

Conclusions:

• PhyloMissForest offers a highly customizable and robust framework for phylogenetic missing data imputation.
• The method achieves significant topological accuracy in inferring evolutionary relationships.
• PhyloMissForest provides effective speedups compared to existing state-of-the-art methods.