A quartet-based approach for inferring phylogenetically informative features from genomic and phenomic data.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Phylogenetics

Background:

• Neural networks are increasingly used in bioinformatics for feature extraction from diverse biological data.
• A critical challenge is ensuring that extracted features align with established evolutionary relationships represented by phylogenetic trees.
• Current methods often lack a principled way to integrate phylogenetic information into feature learning.

Purpose of the Study:

• To develop a machine learning framework that integrates phylogenetic constraints into neural network-based feature extraction.
• To create a latent feature space where pairwise distances reflect the topology of a reference phylogenetic tree.
• To provide a method for validating the evolutionary compatibility of extracted biological features.

Main Methods:

• A novel machine learning approach utilizing neural networks trained with taxon-specific data and a reference phylogenetic tree.
• Development of a quartet-based loss function, leveraging the established role of quartets in phylogenetic analysis.
• Application of the framework to bacterial ribosomal RNA sequences as a proof-of-concept.

Main Results:

• The learned feature distances from the neural network demonstrated high consistency with the reference phylogenetic tree.
• The quartet-based loss function effectively guided the neural network to produce phylogenetically informed feature representations.
• Successful application in a case study using bacterial ribosomal RNA sequences.

Conclusions:

• The proposed framework offers a principled method for incorporating phylogenetic information into neural network feature extraction.
• This approach enhances the evolutionary interpretability of features derived from biological data.
• The method is versatile and applicable to various biological data types for robust phylogenetic analysis.