# A pre-trained language model-based cross-modal fusion framework for predicting miRNA-drug resistance and sensitivity associations

## TL;DR

This paper introduces PLMF-MDA, a computational model using pre-trained language models to predict how microRNAs affect cancer drug resistance and sensitivity, potentially speeding up drug development.

## Contribution

PLMF-MDA is a novel cross-modal fusion framework that integrates pre-trained language models with multi-scale feature extractors to predict miRNA-drug associations.

## Key findings

- PLMF-MDA outperforms existing methods in predicting miRNA-drug resistance and sensitivity associations.
- The model successfully identifies novel miRNA-drug resistance and sensitivity candidates in case studies involving docetaxel and gefitinib.

## Abstract

MicroRNAs (miRNAs) are pivotal regulators of drug resistance and sensitivity in cancer cells, functioning as tumor suppressors or oncogenes that modulate the cellular response to anticancer drugs. While experimental identification of miRNA-mediated drug resistance and sensitivity is both costly and laborious, computational methods present a promising alternative. Recent advances in pre-trained language models (PLMs) offer new opportunities to leverage large-scale unlabeled biomolecular data for enhanced relationship prediction. In this study, we introduce PLMF-MDA, a PLM-based cross-modal fusion model designed to predict miRNA-drug resistance (MDR) and miRNA-drug sensitivity (MDS) associations. PLMF-MDA integrates miRNA and drug multimodal embeddings derived from PLMs and intrinsic feature extractors, and employs a cross-modal attention fusion module to adaptively capture key interactions between modalities. To evaluate the performance of the approach, we manually constructed two benchmark datasets. Experimental results demonstrate that the PLMF-MDA achieves superior prediction performance. Furthermore, case studies on anticancer drug docetaxel and gefitinib demonstrate its potential in discovering novel MDR (MDS) associations. All data and source code are available on GitHub: https://github.com/sheng-n/PLMF-MDA.

MicroRNAs (miRNAs) are important modulators of cancer cell response to chemotherapy, but experimentally identifying miRNA-mediated drug resistance and sensitivity relationships is slow and resource-intensive. Here we present PLMF-MDA, a computational framework that leverages pre-trained language models (RNA-FM and ChemBERTa-2) to generate global embeddings for miRNA sequences and drug SMILES, and combines these with fine-grained nucleotide- and atom-level representations learned by multi-scale CNN and GCN. A cross-modal attention fusion module adaptively integrates these heterogeneous features to capture key interactions between miRNAs and drugs. We curated two benchmark datasets (MDRdataset and MDSdataset) to evaluate the approach, PLMF-MDA consistently outperforms existing methods on AUC and AUPR metrics and maintains strong performance on datasets containing previously unseen nodes. A case study on the anticancer drug docetaxel and gefitinib further demonstrates the model’s ability to prioritize plausible novel miRNA-drug resistance and sensitivity candidates. PLMF-MDA can serve as a practical tool for researchers by identifying miRNAs that may potentiate the efficacy of specific drugs, thereby helping to narrow experimental targets and accelerate the development of more effective cancer therapies. All data and source code are freely available at to facilitate reuse and further research by the community.