Research map / Probabilistic and scientific modeling / Molecules and proteins
Triangle Multiplication is All You Need for Biomolecular Structure Representations
Jeffrey Ouyang-Zhang, Pranav Murugan, Daniel Jesus Diaz, Gianluca Scarpellini, Richard Strong Bowen, Nate Gruver, Adam Klivans, Philipp Kraehenbuehl, Aleksandra Faust, Maruan Al-Shedivat
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 |
|---|---|---|---|
| replace Pairformer with Pairmixer at medium scale | Boltz-1 medium · RCSB test set | lDDT: 0.75 -> 0.76 | Improved |
| replace Pairformer with Pairmixer at large scale | Boltz-1 large · RCSB test set | lDDT: 0.76 -> 0.75 | Hurt |
| replace Pairformer with Pairmixer on GPU time | Boltz-1 large · training GPU-days | GPU-days: 290 -> 192 | Improved |
| replace Pairformer with Pairmixer for inference at 512 tokens | Boltz-1 · sample time on GH200 | seconds: 34 -> 21 | Improved |
| replace sequence-only Transformer with Pairmixer on antibody-antigen | cofolding backbone · Antibody-antigen complex | DockQ>0.23: 0.08 -> 0.23 | Improved |
| replace Pairmixer with sequence-only Transformer on RNA | cofolding backbone · RNA structure | lDDT: 0.59 -> 0.61 | Mixed |
| add triangle multiplication to Pairformer (ablation) | Pairformer small · Pairformer ablation | lDDT: 0.70 -> 0.74 | Improved |
| add triangle attention to Pairformer (ablation) | Pairformer small · Pairformer ablation | lDDT: 0.70 -> 0.74 | Improved |
| add sequence update to Pairformer (ablation) | Pairformer small · Pairformer ablation | lDDT: 0.73 -> 0.74 | No clear effect |
| use triangle multiplication on both directions instead of FFT mixing (ablation) | Pairmixer · mixing method ablation | lDDT: 0.66 -> 0.71 | Improved |
| double FFN hidden dimension from 512 to 1024 (ablation) | Pairmixer · FFN dimension ablation | lDDT: 0.71 -> 0.74 | Improved |
| double triangle multiplication dimension from 128 to 256 (ablation) | Pairmixer · TriMul dimension ablation | lDDT: 0.71 -> 0.73 | Improved |
Closest papers by meaning
- Two Stages of Folding: Convergent Mechanisms in AI Protein Folding Trunks (NeurIPS 2026 · similarity 0.84)
- NeuralPLexer3: Accurate Biomolecular Complex Structure Prediction with Flow Models (NeurIPS 2025 · similarity 0.84)
- 3DMolFormer: A Dual-channel Framework for Structure-based Drug Discovery (ICLR 2025 · similarity 0.84)
- AlphaFold Database Debiasing for Robust Inverse Folding (NeurIPS 2025 · similarity 0.84)
- DualEqui: A Dual-Space Hierarchical Equivariant Network for Large Biomolecules (NeurIPS 2025 · similarity 0.83)
- Evaluating Representation Learning on the Protein Structure Universe (ICLR 2024 · similarity 0.83)
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.
