Research map / Large language models / Agents, code and math
Lost in Tokenization: Context as the Key to Unlocking Biomolecular Understanding in Scientific LLMs
Kai Zhuang, Jiawei Zhang, Yumou Liu, Hanqun Cao, Chunbin Gu, Mengdi Liu, Zhangyang Gao, Zitong Jerry Wang, Xuanhe Zhou, Pheng-Ann Heng, Lijun Wu, Conghui He and 1 more
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 raw protein sequence to the context input (ablation) | Evolla · Protein QA benchmark (All) | LLM-Score: 74.02 -> 70.53 | Hurt |
| add raw protein sequence to the context input (ablation) | Intern-S1 · Protein QA benchmark (All) | LLM-Score: 86.15 -> 84.03 | Hurt |
| add raw protein sequence to the context input (ablation) | Gemini2.5 Pro · Protein QA benchmark (All) | LLM-Score: 87.19 -> 86.98 | Hurt |
| add raw protein sequence to the context input (ablation) | Deepseek-v3 · Protein QA benchmark (All) | LLM-Score: 84.99 -> 86.03 | No clear effect |
| add context to raw sequence input (ablation) | Intern-S1 · Protein QA benchmark (All) | LLM-Score: 43.33 -> 84.03 | Improved |
| add context to raw sequence input (ablation) | NatureLM · Protein QA benchmark (All) | LLM-Score: 6.82 -> 38.86 | Improved |
| add ProTrek output to Pfam + GO context unconditionally (ablation) | DeepSeek context pipeline · Protein QA benchmark (Hard subset) | LLM-Score: 84.60 -> 81.56 | Hurt |
| use ProTrek only as fallback when Pfam and GO are absent (ablation) | DeepSeek context pipeline · Protein QA benchmark (Hard subset) | LLM-Score: 81.56 -> 84.99 | Improved |
| add Pfam to GO context (ablation) | DeepSeek context pipeline · Protein QA benchmark (Hard subset) | LLM-Score: 84.02 -> 84.60 | Improved |
| add ProTrek to GO context (ablation) | DeepSeek context pipeline · Protein QA benchmark (Hard subset) | LLM-Score: 84.02 -> 77.78 | Hurt |
| add ProTrek to Pfam context (ablation) | DeepSeek context pipeline · Protein QA benchmark (Hard subset) | LLM-Score: 74.90 -> 77.00 | Improved |
| replace sequence-as-modality Q-Former alignment with direct textual context | Evolla embeddings · Protein embeddings vs UniClust50 clusters | ARI: 0.809 -> 0.958 | Improved |
Closest papers by meaning
- LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling (NeurIPS 2026 · similarity 0.88)
- Contextualizing biological perturbation experiments through language (ICLR 2025 · similarity 0.88)
- KnowMol: Advancing Molecular Large Language Models with Multi-Level Chemical Knowledge (NeurIPS 2025 · similarity 0.87)
- Bio-xLSTM: Generative modeling, representation and in-context learning of biological and chemical sequences (ICLR 2025 · similarity 0.87)
- Mol-LLaMA: Towards General Understanding of Molecules in Large Molecular Language Model (NeurIPS 2025 · similarity 0.87)
- Generalists vs. Specialists: Evaluating LLMs on Highly-Constrained Biophysical Sequence Optimization Tasks (ICML 2025 · 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.
