Research map / Probabilistic and scientific modeling / Molecules and proteins
BiHiTo: Biomolecular Hierarchy-inspired Tokenization
Ruochong Zheng, Yutian Liu, Yian Zhao, Zhiwei Nie, Xuehan Hou, Chang Liu, Siwei Ma, Youdong Mao, Jie Chen
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 level-3 full C-alpha/C3' hierarchy level (ablation) | BiHiTo · CASP15 | RMSD (A): 0.682 -> 0.585 | Improved |
| add level-3 hierarchy level (ablation) | BiHiTo · RNA3DB test | RMSD (A): 0.622 -> 0.578 | Improved |
| add level-3 hierarchy level (ablation) | BiHiTo · CATH4.2 test | RMSD (A): 0.528 -> 0.512 | Improved |
| add level-2 strided C-alpha sampling level (ablation) | BiHiTo · CASP15 | RMSD (A): 0.646 -> 0.585 | Improved |
| add level-2 strided C-alpha sampling level (ablation) | BiHiTo · CATH4.2 test and RNA3DB test | RMSD (A): 0.527 -> 0.512 (CATH4.2) | No clear effect |
| use multi-codebook hierarchical quantization instead of single-codebook Bio2Token | BiHiTo · FastFolding (out-of-distribution) | RMSD (A): 0.965 -> 0.469 | Improved |
| use hierarchical tokenizer instead of Bio2Token | BiHiTo · CASP14 | RMSD (A): 0.644 -> 0.515 | Improved |
| use hierarchical tokenizer instead of Bio2Token | BiHiTo · RNA3DB test | RMSD (A): 0.771 -> 0.578 | Improved |
| match codebook size to Bio2Token at 4096 codes | BiHiTo · CATH4.2 test | RMSD (A): 0.512 -> 0.499 | No clear effect |
| use hierarchy tokenizer on small molecules (L5 only) | BiHiTo · small molecule benchmark | reconstruction accuracy | No clear effect |
| use BiHiTo versus ProTokens on CASP14 TM-score | BiHiTo · CASP14 | TM-score: 0.991 -> 0.987 | Hurt |
Closest papers by meaning
- Protein Structure Tokenization: Benchmarking and New Recipe (ICML 2025 · similarity 0.87)
- Hierarchical Graph Tokenization for Molecule-Language Alignment (ICML 2025 · similarity 0.85)
- Proteina: Scaling Flow-based Protein Structure Generative Models (ICLR 2025 · similarity 0.85)
- Ensembits: an alphabet of protein conformational ensembles (NeurIPS 2026 · similarity 0.85)
- Few-Step Cofolding with All-Atom Flow Maps (NeurIPS 2026 · similarity 0.85)
- Atomas: Hierarchical Adaptive Alignment on Molecule-Text for Unified Molecule Understanding and Generation (ICLR 2025 · 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.
