DrugKLM: Hybrid AI Framework for Therapeutic Prioritization
DrugKLM represents a hybrid system that combines the structure of biomedical knowledge graphs with mechanistic reasoning derived from large language models, aimed at prioritizing therapies based on their mechanisms. It surpasses both knowledge graph-only and language model-only benchmarks, including TxGNN, on various datasets. The confidence scores from DrugKLM align well with molecular phenotypes, showing that higher scores correlate with transcriptional signatures associated with better survival rates in 12 TCGA cancers. This scoring system effectively identifies biologically perturbational signals over traditional indication trends. Additionally, expert curation in five cancers uncovers consistent patterns.
Key facts
- DrugKLM is a hybrid framework integrating biomedical knowledge graphs and LLMs.
- It outperforms TxGNN and other baselines.
- Confidence scores align with molecular phenotypes across 12 TCGA cancers.
- Higher scores correlate with transcriptional signatures linked to improved survival.
- Framework captures biologically perturbational signals over historical indication patterns.
- Expert curation across five cancers reveals systematic patterns.
Entities
Institutions
- TCGA