Research map / Probabilistic and scientific modeling / Molecules and proteins
Scaling Atomistic Protein Binder Design with Generative Pretraining and Test-Time Compute
Kieran Didi, Zuobai Zhang, Guoqing Zhou, Danny Reidenbach, Zhonglin Cao, Sooyoung Cha, Tomas Geffner, Christian Dallago, Jian Tang, Michael M. Bronstein, Martin Steinegger, Emine Kucukbenli and 2 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 Teddymer synthetic dimer data to training (ablation) | Complexa · protein binder benchmark | unique successes | Improved |
| add global translation noise to the flow-matching objective (ablation) | Complexa · protein binder benchmark | unique successes / binder placement | Improved |
| add f_ipAE reward to beam search (ablation) | Complexa beam search · protein targets | # unique successes: 77.00 -> 83.36 | Improved |
| add interface H-bond reward to beam search (ablation) | Complexa beam search · protein targets | avg # interface H-bonds: 5.271 -> 7.154 | Improved |
| combine f_ipAE and f_H-Bond rewards (ablation) | Complexa beam search · protein targets | # unique successes: 83.36 -> 86.26 | Improved |
| use structured search (beam search, FKS, MCTS) instead of best-of-N (ablation) | Complexa inference-time scaling · hard protein targets | unique success rate vs GPU hours | Improved |
| use simple best-of-N instead of structured search (ablation) | Complexa inference-time scaling · easy protein targets | unique success rate vs GPU hours | No clear effect |
| initialize BindCraft from Complexa samples (G&H) | BindCraft · easy protein targets | search speed | Improved |
| initialize BindCraft from Complexa samples (G&H) | BindCraft · hard protein targets | search speed | No clear effect |
| add ProteinMPNN re-design of sequences (ablation) | Complexa · protein targets | avg # unique successes: 9.10 -> 14.4 | Improved |
| fix interface residues during MPNN re-design (ablation) | Complexa · protein targets | avg # unique successes: 13.6 -> 14.4 | Mixed |
| extend inference-time search horizon beyond 100 GPU hours | Complexa · TNF-alpha, H1, IL17A | unique successes: 15 for TNF-alpha, 7 for H1, 1 for IL17A | Improved |
Closest papers by meaning
- CBGBench: Fill in the Blank of Protein-Molecule Complex Binding Graph (ICLR 2025 · similarity 0.88)
- BindGPT: A Scalable Framework for 3D Molecular Design via Language Modeling and Reinforcement Learning (AAAI 2025 · similarity 0.88)
- UniMoMo: Unified Generative Modeling of 3D Molecules for De Novo Binder Design (ICML 2025 · similarity 0.88)
- Proteina: Scaling Flow-based Protein Structure Generative Models (ICLR 2025 · similarity 0.88)
- Latent Retrieval Augmented Generation of Cross-Domain Protein Binders (NeurIPS 2025 · similarity 0.87)
- An All-Atom Generative Model for Designing Protein Complexes (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.
