Research map / Large language models / Agents, code and math
SpectraLLM: Uncovering the Ability of LLMs for Molecule Structure Elucidation from Multi-Spectra
Yunyue Su, Jiahui Chen, Zao Jiang, Zhenyi Zhong, Liang Wang, Qiang Liu, Zhaoxiang Zhang
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 |
|---|---|---|---|
| fuse IR, Raman and UV-Vis spectra in one prompt instead of single spectrum (ablation) | SpectraLLM · QM9s | Tanimoto (ECFP4): 0.2500 -> 0.3355 (Raman alone vs joint) | Improved |
| fuse IR, Raman and UV-Vis spectra (ablation) | SpectraLLM · QM9s | MCES: more than 10 -> 4.9647 | Improved |
| fuse 1H, 13C and HSQC NMR instead of one NMR type (ablation) | SpectraLLM · Multimodal Spectroscopic | Tanimoto (ECFP4): 0.2058 (HSQC alone) -> 0.4151 | Improved |
| add IR and MS to NMR input (ablation) | SpectraLLM · Multimodal Spectroscopic | Tanimoto (ECFP4): 0.4151 -> 0.4875 | Improved |
| add IR to NMR input (without MS) (ablation) | SpectraLLM · Multimodal Spectroscopic | Tanimoto (ECFP4): 0.4151 -> 0.4121 | No clear effect |
| add MS to IR input (ablation) | SpectraLLM · Multimodal Spectroscopic | Tanimoto (ECFP4): 0.1844 (MS) / 0.1720 (IR) -> 0.2300 | Improved |
| use LLM text-based spectra representation vs specialized baselines on NMR | SpectraLLM vs NMR2Struct · Multimodal Spectroscopic NMR | Tanimoto (ECFP4): 0.0433 -> 0.4151 | Improved |
| use SpectraLLM vs baselines on UV-Vis alone | SpectraLLM · QM9s UV-Vis | Tanimoto (ECFP4): 0.0965/0.0716 -> 0.0790 (modest) | Improved |
| use SpectraLLM vs DiffMS on MassSpecGym | SpectraLLM · MassSpecGym | Tanimoto (ECFP4): 0.1597 (DiffMS) vs 0.1533 | No clear effect |
| use SpectraLLM vs baselines on MassBank | SpectraLLM · MassBank | functional group overlap: 0.2999 -> 0.4539 | Improved |
| train on multi-spectral inputs then test on unseen single-spectrum (ablation) | SpectraLLM · QM9s (cross-domain) | generalization | Improved |
| replace text-based LLM with vision-language models (ablation) | SpectraLLM · structure elucidation | performance | Hurt |
Closest papers by meaning
- Can LLMs Solve Molecule Puzzles? A Multimodal Benchmark for Molecular Structure Elucidation (NeurIPS 2024 · similarity 0.90)
- Unraveling Molecular Structure: A Multimodal Spectroscopic Dataset for Chemistry (NeurIPS 2024 · similarity 0.89)
- MS-BART: Unified Modeling of Mass Spectra and Molecules for Structure Elucidation (NeurIPS 2025 · similarity 0.87)
- RFL: Simplifying Chemical Structure Recognition with Ring-Free Language (AAAI 2025 · similarity 0.86)
- ChemVLM: Exploring the Power of Multimodal Large Language Models in Chemistry Area (AAAI 2025 · similarity 0.86)
- Domain-Agnostic Molecular Generation with Chemical Feedback (ICLR 2024 · similarity 0.85)
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.
