Large Language Models Meet Biomedical Knowledge Graphs for Mechanistically Grounded Therapeutic Prioritization
Summary: arXiv:2604.19815v1 Announce Type: new
Abstract: Drug repurposing is often framed as a candidate identification task, but existing approaches provide limited guidance for distinguishing biologically plausible candidates from historically well-connected ones. Here we introduce DrugKLM, a hybrid framework that integrates biomedical knowledge graph structure with large language model-based mechanistic reasoning to enable mechanistically grounded therapeutic prioritization. Across benchmark datasets, DrugKLM outperforms knowledge graph-only and language model-only baselines, including TxGNN. Beyond improved recall, DrugKLM confidence scores exhibit functional alignment with molecular phenotypes: higher scores are associated with transcriptional signatures linked to improved survival across 12 TCGA cancers. The scoring framework preferentially captures biologically perturbational signals rather than historical indication patterns. Expert curation across five cancers further reveals systematic differences in prioritization behavior, with DrugKLM elevating candidates supported by coherent mechanistic rationale and disease-specific clinical context. Together, these results establish DrugKLM as an evidence-integrative framework that translates heterogeneous biomedical data into mechanistically interpretable and clinically grounded therapeutic hypotheses.
Introduction
The landscape of drug discovery is rapidly evolving, particularly with the advent of artificial intelligence (AI) and machine learning technologies. Drug repurposing, which involves identifying new uses for existing drugs, has gained traction as a cost-effective strategy in therapeutic development. However, distinguishing between biologically plausible candidates and those that are historically connected remains a challenge. This article discusses the innovative DrugKLM framework that utilizes large language models (LLMs) integrated with biomedical knowledge graphs to enhance the candidate prioritization process.
DrugKLM: A New Paradigm in Drug Repurposing
DrugKLM stands out as a hybrid framework that synthesizes the strengths of knowledge graphs and language models. The primary goals of this framework include:
- Enhancing the identification of therapeutically relevant candidates through mechanistic reasoning.
- Improving recall rates in drug candidate identification compared to traditional methods.
- Aligning confidence scores with molecular phenotypes to predict clinical outcomes.
Performance and Validation
DrugKLM has been benchmarked against various existing approaches, including TxGNN, a leading knowledge graph-based method. The results demonstrate a marked improvement in performance metrics. Key findings include:
- DrugKLM consistently outperformed knowledge graph-only and language model-only baselines.
- Higher confidence scores were found to correlate with transcriptional signatures indicative of better survival rates across 12 cancers as outlined by The Cancer Genome Atlas (TCGA).
Mechanistic Insights and Clinical Relevance
One of the most significant advantages of DrugKLM is its ability to capture biologically perturbational signals. This is crucial as it shifts the focus from historical patterns of drug indications to a more nuanced understanding of mechanistic relationships within the biological context. Expert curation of five different cancers revealed:
- DrugKLM prioritizes candidates that are supported by coherent mechanistic rationale.
- It takes into account disease-specific clinical contexts, providing a more comprehensive approach to therapeutic prioritization.
Conclusion
In conclusion, DrugKLM represents a significant advancement in the integration of AI with biomedical research. By combining large language models with knowledge graphs, it not only enhances drug repurposing strategies but also establishes a framework for generating clinically relevant therapeutic hypotheses. This innovative approach promises to contribute significantly to the future of personalized medicine and drug discovery, offering hope for more effective treatments across various diseases.