Phylogenetic tree information aids supervised learning for predicting protein-protein interaction based on distance...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular processes.
• Predicting PPIs computationally often relies on co-evolutionary information from orthologous proteins across different species.
• Existing methods analyze correlations within distance matrices derived from orthologs.

Purpose of the Study:

• To develop a novel and simple method for improving PPI prediction accuracy.
• To leverage phylogenetic information to better capture intra-matrix correlations in ortholog distance matrices.

Main Methods:

• Utilized a phylogenetic species tree as a guide for hierarchical clustering of orthologous proteins.
• Generated intermediate distance matrices from cluster distances using the Neighbor Joining algorithm.
• Transformed and concatenated these matrices into a super phylogenetic vector for machine learning.
• Trained a support vector machine (SVM) on known PPIs represented by these vectors.

Main Results:

• The proposed method achieved a significantly higher prediction accuracy (ROC score of 0.8446) compared to using Pearson correlations (ROC score of 0.6587).
• Cross-validation experiments demonstrated the effectiveness of the phylogenetic approach.

Conclusions:

• Phylogenetic trees effectively guide the extraction of intra-matrix correlations from ortholog distance matrices.
• The inclusion of these intermediate matrices, derived from ancestral orthologs, improves both unsupervised and supervised learning paradigms for PPI prediction.
• The method offers a superior balance between sensitivity and specificity in predicting protein-protein interactions.