Research map / Large language models / Instruction and fine-tuning
Entropy-Guided Dynamic Tokens for Graph-LLM Alignment in Molecular Understanding
Zihao Jing, QIUHAO Zeng, Ruiyi Fang, Yan Sun, Boyu Wang, Pingzhao Hu
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 |
|---|---|---|---|
| add entropy-guided patching instead of BRICS fragments (ablation) | EDT-Former · Pampa | accuracy (%): 71.90 -> 84.52 | Improved |
| add entropy-guided patching instead of random patches (ablation) | EDT-Former · BBBP | accuracy (%): 68.90 -> 75.06 | Improved |
| add any patching (entropy) versus none (ablation) | EDT-Former · BBBP | accuracy (%): 39.67 -> 75.06 | Improved |
| add entropy-guided patching versus BRICS on BBBP (ablation) | EDT-Former · BBBP | accuracy (%): 73.59 -> 75.06 | Improved |
| add modality fusion (text, graph, 3D alignment) to the connector (ablation) | EDT-Former · MoleculeQA | average accuracy: 26% avg. degradation when removed | Improved |
| add Dynamic Query Transformer (dynamic tokens) instead of fixed-length projection (ablation) | EDT-Former · MoleculeQA | average accuracy: 10.7% average drop when removed | Improved |
| add Entropy-Guided Patching instead of uniform 3-token SMILES patches (ablation) | EDT-Former · MoleculeQA | average accuracy: 11.5% average drop when removed | Improved |
| add the 3D modality to text and graph (ablation) | EDT-Former · MoleculeQA | total accuracy (%): 65.11 -> 68.34 | Improved |
| add the graph modality to text and 3D (ablation) | EDT-Former · MoleculeQA | total accuracy (%): 60.48 -> 68.34 | Improved |
| add graph and 3D inputs to text-only SMILES (ablation) | EDT-Former · MoleculeQA | total accuracy (%): 58.77 -> 68.34 | Improved |
| attach EDT-Former connector to Mistral-8B | Mistral-8B · MoleculeQA | total accuracy (%): 28.00 -> 55.63 | Improved |
| attach EDT-Former connector to Llama3.1-8B | Llama3.1-8B · MoleculeQA | total accuracy (%): 49.31 -> 68.34 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Mol-LLaMA-2 — Mol-LLaMA: Towards General Understanding of Molecules in Large Molecular Language Model (NeurIPS 2025) · better on 6 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Bridging Molecular Graphs and Large Language Models (AAAI 2025 · similarity 0.93)
- LLaMo: Large Language Model-based Molecular Graph Assistant (NeurIPS 2024 · similarity 0.91)
- Hierarchical Graph Tokenization for Molecule-Language Alignment (ICML 2025 · similarity 0.91)
- Atomas: Hierarchical Adaptive Alignment on Molecule-Text for Unified Molecule Understanding and Generation (ICLR 2025 · similarity 0.88)
- Improving Large Molecular Language Model via Relation-aware Multimodal Collaboration (AAAI 2026 · similarity 0.88)
- KnowMol: Advancing Molecular Large Language Models with Multi-Level Chemical Knowledge (NeurIPS 2025 · similarity 0.88)
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.
