Research map / Probabilistic and scientific modeling / Molecules and proteins
FlexProtein: Joint Sequence and Structure Pretraining for Protein Modeling
Jianwei Zhu, Yu Shi, Ran Bi, Peiran Jin, Chang Liu, Zhe Zhang, Haitao Huang, Zekun Guo, Pipi Hu, Fusong Ju, Lin Huang, Xinwei Tai and 11 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 |
|---|---|---|---|
| joint sequence-structure pretraining (MLM plus diffusion) instead of MSA-based or sequence-only models | FlexRibbon · SAb-23H2 antibody structure prediction | success rate: 0.4667 -> 0.6133 vs IgGM | Improved |
| joint sequence-structure pretraining | FlexRibbon · FoldBench protein-peptide | success rate: 0.8431 -> 0.9137 vs AlphaFold 3 | Improved |
| joint sequence-structure pretraining for nanobody CDR design | FlexRibbon · Nanobody CDR design | DockQ: 0.267 -> 0.244 DockQ; 0.183 -> 0.218 AAR | Mixed |
| add ligand context when predicting conformational ensembles (ablation) | FlexRibbon · Apo/Holo conformational change | TM-ens: +0.012 (0.889 baseline mode) | Improved |
| joint structure pretraining versus sequence-only encoders for function prediction | FlexRibbon · GO-BP, GO-MF, GO-CC F1 | F1: +0.051, +0.013, +0.096 over ESM-GearNet | Improved |
| joint pretraining versus ESM-GearNet on EC | FlexRibbon · EC | F1: 0.890 -> 0.891 | No clear effect |
| joint pretraining for protein-ligand docking versus single-sequence baselines | FlexRibbon · PoseBusters V1 | success rate: 71.82% random-1, 78.70% oracle | Improved |
| joint pretraining for binding affinity | FlexRibbon · CASF-2016 | RMSE: 1.258 -> 1.150 | Improved |
| remove alignment-based loss (use plain denoising with random rotation augmentation) (ablation) | FlexRibbon diffusion · pretraining | training stability | No clear effect |
| defer MLM loss to stage B (ablation) | FlexRibbon pretraining · pretraining | stability | Improved |
Closest papers by meaning
- Learning to engineer protein flexibility (ICLR 2025 · similarity 0.91)
- DPLM-2: A Multimodal Diffusion Protein Language Model (ICLR 2025 · similarity 0.90)
- Diffusion on Language Model Encodings for Protein Sequence Generation (ICML 2025 · similarity 0.89)
- Diffusion Language Models Are Versatile Protein Learners (ICML 2024 · similarity 0.89)
- ProSST: Protein Language Modeling with Quantized Structure and Disentangled Attention (NeurIPS 2024 · similarity 0.89)
- FlexMol: A Flexible Toolkit for Benchmarking Molecular Relational Learning (NeurIPS 2024 · similarity 0.88)
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.
