Research map / Large language models / Retrieval and knowledge
Knowledge Graph and Large Language Model for Metabolomics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
A doctoral consortium-style research statement summarizing the author's work on a metabolite-centric knowledge graph (MetaKG), multimodal clinical data integration, and KG-enhanced LLMs, and outlining future plans for metabolomics.
Problem
Lack of tools for integrating, analyzing, and inferring diverse metabolomics knowledge, and the unexplored synergy between knowledge graphs and LLMs for metabolomics.
Approach
Summarizes a metabolite-centric knowledge graph (MetaKG) built from HMDB with KG representation learning and a triple contrastive learning module, plus prior methods that inject heterogeneous knowledge into LLM prompts; proposes future MetaKG-LLM synergy via RAG and a metabolomics domain LLM.
Key results
- Multimodal integration of DNA methylation, mRNA, and miRNA data reached an average AUC of 87.41% across 12 cancer subtype prediction tasks, versus prior SOTA 84.85%.
- MetaKG achieves state-of-the-art results on 12 metabolite attribute prediction tasks and on a NAFLD clinical diagnosis task (as reported by the author, from prior work).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use dynamic learning and guided cross-attention to integrate multimodal data | Multimodal cancer subtype prediction · 12 cancer subtype prediction tasks | average AUC: 84.85% -> 87.41% | Helped |
| Use MetaKG representation learning and multimodal metabolite representation integration with triple contrastive learning | MetaKG · Metabolite attribute prediction tasks | prediction results | Helped |
| Use MetaKG representation learning and multimodal metabolite representation integration with triple contrastive learning | MetaKG · NAFLD clinical diagnosis task | diagnosis results | Helped |
| Convert heterogeneous external knowledge to unified natural language and use prompt learning or LLM agents | LLM for EHR diagnostic classification · EHR diagnostic classification dataset | diagnosis and cross-departmental transfer results | Helped |
Limitations
- Position-style summary with no new experiments; results are reported from prior work by the author.
- Future directions are proposals without reported experiments.
Experimental setup
Datasets: HMDB
Metrics: AUC
Builds on: retrieval-augmented generation
Closest papers by meaning
- Enhancing Aging Biomarker Research through Large Language Models and Knowledge Graphs (Student Abstract) (AAAI 2025 · similarity 0.85)
- NutriBench: A Dataset for Evaluating Large Language Models in Nutrition Estimation from Meal Descriptions (ICLR 2025 · similarity 0.81)
- MASS: Mathematical Data Selection via Skill Graphs for Pretraining Large Language Models (ICML 2025 · similarity 0.81)
- How Do Large Language Models Understand Graph Patterns? A Benchmark for Graph Pattern Comprehension (ICLR 2025 · similarity 0.81)
- MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search (NeurIPS 2025 · similarity 0.81)
- K-ON: Stacking Knowledge on the Head Layer of Large Language Model (AAAI 2025 · similarity 0.80)
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.