An Inductive Logistic Matrix Factorization Model for Predicting Drug-Metabolite Association With Vicus...

Area of Science:

• Pharmacology
• Computational Biology
• Bioinformatics

Background:

• Metabolites are closely linked to human diseases.
• Drug-metabolite interactions are crucial in pharmacomicrobiomics but are challenging to study experimentally.
• Existing computational methods have not extensively predicted large-scale drug-metabolite associations.

Purpose of the Study:

• To develop a novel algorithm, inductive logistic matrix factorization (ILMF), for predicting latent drug-metabolite associations.
• To integrate various interaction data (drug-drug, metabolite-metabolite, drug-metabolite) into a unified prediction framework.
• To leverage inductive matrix completion and fused feature representations for enhanced prediction accuracy.

Main Methods:

• Proposed the inductive logistic matrix factorization (ILMF) algorithm.
• Integrated drug-drug, metabolite-metabolite, and drug-metabolite interaction data.
• Utilized inductive matrix completion with low-dimensional feature representations (F and F) for learning projection matrices (U and V).
• Employed the Vicus spectral matrix to encode refined local geometrical structures.

Main Results:

• ILMF demonstrated competitive performance against state-of-the-art approaches.
• The method effectively predicts potential drug-metabolite associations.
• Experimental validation was performed on the "DrugMetaboliteAtlas" dataset.

Conclusions:

• ILMF is an effective computational approach for predicting drug-metabolite associations.
• The method addresses the limitations of experimental validation in pharmacomicrobiomics.
• This work contributes to a larger-scale understanding of drug-metabolite interactions.