Research map / Probabilistic and scientific modeling / Molecules and proteins
Property-Driven Protein Inverse Folding with Multi-Objective Preference Alignment
Junqi Liu, Xiaoyang Hou, Chence Shi, Xin Liu, Zhi Yang, Jian Tang
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 |
|---|---|---|---|
| fine-tune ProteinMPNN with multi-objective semi-online DPO (MoMPNN Sol+TM) | ProteinMPNN · CATH 4.3 test | Solubility (Sol): 0.719 -> 0.884 | Improved |
| fine-tune ProteinMPNN with multi-objective DPO (MoMPNN Thermo+IG) | ProteinMPNN · CATH 4.3 test | Thermostability (Thermo): 0.769 -> 0.963 | Improved |
| fine-tune with multi-objective DPO (Sol+TM) | ProteinMPNN · CATH 4.3 test | TM score: 0.740 -> 0.738 | No clear effect |
| use IG objective instead of TM for solubility alignment (ablation) | MoMPNN · CATH 4.3 test | RMSD: 4.37 -> 4.73 | Hurt |
| use IG instead of TM as designability objective (ablation) | MoMPNN · De novo backbones | TM score: 0.734 -> 0.742 (Sol) | Improved |
| add EP objective to Sol+TM (ablation) | MoMPNN · CATH 4.3 test | Solubility (Sol): 0.884 -> 0.852 | Hurt |
| add EP objective to Sol+TM (ablation) | MoMPNN · CATH 4.3 test | Thermostability (Thermo): 0.747 -> 0.790 | Improved |
| MoMPNN versus ProteinMPNN on de novo backbones | ProteinMPNN · De novo backbones | TM score: 0.718 -> 0.751 | Improved |
| train on property subset (HyperMPNN) rather than alignment | ProteinMPNN · CATH 4.3 test | TM score: 0.740 -> 0.706 | Hurt |
| multi-objective flexible-margin DPO instead of weighted-score DPO (ablation) | MoMPNN · CATH 4.3 small validation | Evo. ppl: 5.843 (weighted) vs 6.205 (MoMPNN) | Mixed |
| multi-objective flexible-margin DPO instead of weighted-score DPO (ablation) | MoMPNN · CATH 4.3 small validation | TM score: 0.811 -> 0.825 | Improved |
| MoMPNN versus ProteinMPNN on binder design | ProteinMPNN · binder design | success rate | Improved |
Closest papers by meaning
- Protein Inverse Folding From Structure Feedback (NeurIPS 2025 · similarity 0.93)
- ProteinOPD: Towards Effective and Efficient Preference Alignment for Protein Design (NeurIPS 2026 · similarity 0.90)
- ProtInvTree: Deliberate Protein Inverse Folding with Reward-guided Tree Search (NeurIPS 2025 · similarity 0.89)
- Fast Uncovering of Protein Sequence Diversity from Structure (ICLR 2025 · similarity 0.89)
- Bridge-IF: Learning Inverse Protein Folding with Markov Bridges (NeurIPS 2024 · similarity 0.88)
- Learning to engineer protein flexibility (ICLR 2025 · 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.
