Research map / Large language models / Retrieval and knowledge
GALAX: Graph-Augmented Language Model for Explainable Reinforcement-Guided Subgraph Reasoning in Precision Medicine
Heming Zhang, Di Huang, Wenyu Li, Michael A Province, Yixin Chen, Philip Payne, Fuhai Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| task-adaptive finetuning on Target-QA instead of no finetuning | LLaMA3-8B + Omics · Target-QA (TCGA, overall) | precision: 0.0071 -> 0.5250 | Improved |
| domain finetuning on biomedical text instead of none | LLaMA3-8B + Omics · Target-QA (TCGA, overall) | precision: 0.0071 -> 0.0179 | Improved |
| add static KG to task-finetuned LLM | L3-FT(QA) + Omics · Target-QA (TCGA, overall) | precision: 0.5250 -> 0.5185 | No clear effect |
| use GALAX reinforcement-guided subgraph reasoning instead of static KG | L3-FT(QA) + Omics + KG · Target-QA (TCGA, overall) | precision: 0.5185 -> 0.5472 | Improved |
| use GALAX instead of GNN-RAG | graph RAG baseline · Target-QA (TCGA, overall) | recall: 0.4735 -> 0.5332 | Improved |
| replace LLaMA3-8B backbone with Qwen2.5-7B | GALAX · Target-QA (TCGA, overall) | precision: 0.5472 -> 0.5445 | No clear effect |
| add epigenomic data input (ablation) | GALAX · Target-QA (TCGA, overall) | precision: 0.5163 -> 0.5472 | Improved |
| add genomic data input (ablation) | GALAX · Target-QA (TCGA, overall) | precision: 0.4277 -> 0.5472 | Improved |
| add transcriptomic data input (ablation) | GALAX · Target-QA (TCGA, overall) | precision: 0.4065 -> 0.5472 | Improved |
| add proteomic data input (ablation) | GALAX · Target-QA (TCGA, overall) | precision: 0.3980 -> 0.5472 | Improved |
| keep all omics instead of removing all (ablation) | GALAX · Target-QA (TCGA, overall) | precision: 0.3793 -> 0.5472 | Improved |
| keep all edges instead of dropping 20% of KG edges (ablation) | GALAX · Target-QA (TCGA, overall) | recall: 0.5061 -> 0.5332 | Improved |
Closest papers by meaning
- OgBench: A Framework for Evaluating Graph Neural Networks on Omics Data (NeurIPS 2026 · similarity 0.85)
- $\texttt{G1}$: Teaching LLMs to Reason on Graphs with Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- KINDLE: Knowledge-Guided Distillation for Prior-Free Gene Regulatory Network Inference (NeurIPS 2025 · similarity 0.84)
- Beyond Graphs: Can Large Language Models Comprehend Hypergraphs? (ICLR 2025 · similarity 0.84)
- On the Scalability of GNNs for Molecular Graphs (NeurIPS 2024 · similarity 0.84)
- GeSubNet: Gene Interaction Inference for Disease Subtype Network Generation (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
