Graph convolution for predicting associations between miRNA and drug resistance.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNA (miRNA) therapeutics show promise but are hindered by drug resistance.
• Aberrant miRNA expression is linked to drug resistance, yet predictive tools are lacking.
• Understanding miRNA-drug resistance associations is crucial for effective therapeutic development.

Purpose of the Study:

• To develop a computational approach for predicting associations between microRNAs and drug resistance.
• To address the current lack of computational tools for identifying miRNA-drug resistance relationships.
• To facilitate the advancement of miRNA-based therapeutics by predicting drug resistance mechanisms.

Main Methods:

• Formulated the prediction problem as a link prediction task on a bipartite attributed graph.
• Employed graph convolution techniques to build a latent factor model for predicting miRNA-drug resistance associations.
• Integrated high-dimensional miRNA expression profiles, drug substructure fingerprints, and ontology data.

Main Results:

• The GCMDR model achieved high predictive performance with AUCs ranging from 0.9301 to 0.9369 across cross-validation folds.
• Successfully demonstrated the reliable prediction of miRNA-drug resistance associations using integrated biological and chemical data.
• Learned effective graph embedding features for both miRNAs and drugs.

Conclusions:

• The GCMDR computational approach reliably predicts microRNA-drug resistance associations.
• This tool can aid in overcoming drug resistance challenges in miRNA-based therapeutics.
• The integration of diverse data sources enhances the prediction accuracy of miRNA-drug interactions.