Ligand Similarity Complements Sequence, Physical Interaction, and Co-Expression for Gene Function Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Protein-ligand databases enable networking proteins by ligand similarity.
• Ligand-based protein networks complement gene networks for predicting functional relevance.
• Existing functional genomic associations include sequence similarity, protein interactions, co-expression, and disease annotations.

Purpose of the Study:

• To quantify how ligand-based protein associations complement functional genomic associations.
• To assess the added value of integrating ligand-based and functional genomic networks.
• To investigate improvements in gene function prediction through network combination.

Main Methods:

• Constructed a protein network using the Similarity Ensemble Approach (SEA) based on ligand similarity.
• Calculated network similarity across functional genomic associations for 1,131 genes.
• Evaluated Molecular Function prediction within Gene Ontology (GO) using individual and combined networks.

Main Results:

• Ligand-based and functional genomic networks showed small but significant overlaps.
• Combining networks substantially improved Molecular Function prediction in GO (AUROC from ~0.63-0.75 to ~0.8).
• The study demonstrated a boost in guilt-by-association gene function prediction upon network integration.

Conclusions:

• Ligand-based protein networks provide distinct information complementary to functional genomic networks.
• Integration of diverse biological networks enhances predictive power for gene function.
• The findings highlight the potential for exploiting combined network properties in biological discovery.