Towards rational discovery of combination therapies: LLM-enhanced gene function mapping to explainable drug synergy prediction

The National Institute of Environmental Health Sciences awarded the University of Texas at Arlington $632,466 on August 6, 2026, under the Medical Library Assistance program (CFDA 93.879) to develop a computational framework for predicting gene-gene and drug-drug synergies in precision medicine.

The project, "Towards Rational Discovery of Combination Therapies: LLM-Enhanced Gene Function Mapping to Explainable Drug Synergy Prediction," will advance three primary aims: develop multi-modal deep learning approaches to enhance gene function prediction and analyze gene synergies; design BaySyn, a Bayesian hierarchical model integrating combinational screening data, multi-omics data, chemical properties, drug and target gene interactions, and gene-gene synergies to predict drug synergies with explainable mechanisms; and validate predictions through high-throughput screening experiments. The work addresses the current computational gap in integrating diverse datasets comprehensively and providing interpretable predictions, leveraging transformer-based deep learning, large language models, and Bayesian modeling to overcome experimental inefficiencies in identifying synergies.

Work is performed in Arlington, Texas, with a period of performance through June 30, 2031.

Abstract

We aim to develop a cutting-edge computational framework to predict gene-gene and drug-drug synergies, addressing critical gaps in precision medicine. Gene-gene synergies amplify biological effects through gene interactions, while drug-drug synergies enhance therapeutic outcomes in combination therapies. Experimentally identifying such synergies is costly, labor-intensive, and inefficient. Current computational methods, though promising, fall short in integrating diverse datasets comprehensively and providing interpretable predictions-challenges we aim to overcome through innovative approaches. Recent breakthroughs in machine learning (ML) and data integration offer unprecedented opportunities. First, transformer-based deep learning (DL) and large language models (llms) unlock hidden insights into gene functions from vast, multi-modal, unstructured data. Second, bayesian modeling excels in integrating diverse data sources for robust and interpretable synergy predictions. Third, the availability of high-throughput drug screening and gene function databases offers a tangible path to bridge gene-gene and drug-drug synergies for therapeutic applications. Building on these advancements and leveraging our team's unique expertise, we propose three aims: (1) develop advanced multi-modal DL approaches to enhance gene function prediction and provide a robust foundation for comprehensive analysis of gene synergies, addressing the challenge of limited functional annotations for thousands of genes; (2) design baysyn, a bayesian hierarchical model to predict drug synergies with explainable mechanisms by integrating combinational screening data, multi-omics data, chemical properties, drug and target gene interactions, and gene-gene synergies; (3) validate predictions through high-throughput screening experiments, establishing real-world evidence of computational insights. We will also create a public web portal to disseminate our tools and findings, fostering collaboration and accessibility. Our multidisciplinary team, spanning ML, statistics, bioinformatics, and drug development, has a proven track record of innovation. Our prior work includes pioneering DL methods for biological data and bayesian models for integrating complex datasets, laying the foundation for the success of this project. Our study has transformative potential. By combining DL, bayesian modeling, and experimental validation, we aim to accelerate the discovery of safer, more effective combination therapies, reducing the development time and cost. By addressing challenges in interpretability and data integration, we strive to set a new benchmark for synergy prediction. Our public web portal will provide researchers with validated tools and data, driving further advancements in combination therapies and therapeutic target discovery while advancing methods for analyzing complex biomedical data. Together, these advances will shape the future of precision medicine, delivering actionable solutions for unmet clinical needs.