Research map / Large language models / Language, safety and interpretability
What can Large Language Models do in chemistry? A comprehensive benchmark on eight tasks
Taicheng Guo, Kehan Guo, Bozhao Nan, Zhenwen Liang, Zhichun Guo, Nitesh V Chawla, Olaf Wiest, Xiangliang 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 |
|---|---|---|---|
| scaffold-similarity ICL examples instead of random (ablation) | GPT-4 k=20 · reaction prediction | top-1 accuracy: 0.012 -> 0.230 | Improved |
| scaffold-similarity ICL examples instead of random (ablation) | GPT-4 k=8 · BBBP | F1: 0.469 -> 0.587 | Improved |
| scaffold-similarity ICL examples instead of random (ablation) | GPT-4 k=8 · HIV | F1: 0.994 -> 0.797 | Hurt |
| more ICL examples (k=4 -> 8) (ablation) | GPT-4 random · Buchwald-Hartwig yield | accuracy: 0.574 -> 0.800 | Improved |
| more ICL examples (k=4 -> 8) (ablation) | GPT-4 scaffold · BBBP | F1: 0.498 -> 0.587 | Improved |
| more ICL examples (k=5 -> 20) (ablation) | GPT-4 scaffold · reaction prediction | top-1 accuracy: 0.182 -> 0.230 | Improved |
| few-shot ICL instead of zero-shot (ablation) | GPT-4 · Buchwald-Hartwig yield | accuracy: 0.322 -> 0.800 | Improved |
| GPT-4 ICL vs trained Chemformer | reaction prediction · USPTO-MIT | top-1 accuracy: 0.938 -> 0.230 | Hurt |
| GPT-4 ICL vs XGBoost | property prediction · ClinTox | F1: 0.620 -> 0.924 | Improved |
| GPT-4 ICL vs XGBoost | property prediction · BBBP | F1: 0.897 -> 0.587 | Hurt |
| SELFIES instead of SMILES input (ablation) | LLMs · property/reaction/design/captioning | task metrics | Hurt |
| vary temperature 0.2 -> 0.4 (ablation) | GPT-4 scaffold k=8 · BBBP | F1: 0.667 -> 0.712 | Improved |
Closest papers by meaning
- ChemVLM: Exploring the Power of Multimodal Large Language Models in Chemistry Area (AAAI 2025 · similarity 0.91)
- Structured Chemistry Reasoning with Large Language Models (ICML 2024 · similarity 0.91)
- Training a Scientific Reasoning Model for Chemistry (NeurIPS 2025 · similarity 0.89)
- Tag-LLM: Repurposing General-Purpose LLMs for Specialized Domains (ICML 2024 · similarity 0.87)
- CURIE: Evaluating LLMs on Multitask Scientific Long-Context Understanding and Reasoning (ICLR 2025 · similarity 0.87)
- Can LLMs Solve Molecule Puzzles? A Multimodal Benchmark for Molecular Structure Elucidation (NeurIPS 2024 · similarity 0.87)
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.
