SPINET: Sheaf Protein Inverse Folding Network
Abstract
Proteins change shape as they function, yet most inverse folding models predict amino acid sequences from a single, fixed backbone. A central challenge in protein engineering is to design proteins that undergo specific motions, which requires accounting for how their structures change over time. This motivates inverse protein folding conditioned on protein motion. We introduce SPINET, which predicts sequences from molecular dynamics trajectories. It uses cellular sheaves to represent residue interactions within each frame and recurrent units to integrate information across frames, then predicts all amino acids in a single pass. We evaluate SPINET on mdCATH and ATLAS, where it outperforms all evaluated static and ensemble baselines in sequence recovery. On mdCATH, it achieves 56.7% top-1 recovery, compared with 44.5% for the strongest static baseline and 40.7% for the strongest ensemble baseline. We also evaluate whether the predicted sequences are compatible with conformations sampled along the target trajectory. On mdCATH, they achieve a median TM-score of 0.760, and structural recovery favors target conformations over unrelated decoys for 99.5% of test domains.
Related
- An inverse-protein-folding pre-trained model for predicting the mutational effects in protein-protein interactions
- PSLL: Persistent Sheaf Laplacian Learning for Protein-Ligand Binding Affinity Prediction
- Protein Folding Interference Enables Selective Cyclin D1 Degradation
- A passive protein environment offsets RNA folding energetics in cells
- Quasi-continuous cotranslational compaction and folding of a multidomain protein
BibTeX
@misc{lundsgaard2026spinet,
title = {{SPINET: Sheaf Protein Inverse Folding Network}},
author = {Jens Lundsgaard and Colin Mikulski and Zhixuan Yan and Dhananjay Bhaskar},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.34153},
eprint = {2609.34153},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.34153},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - SPINET: Sheaf Protein Inverse Folding Network
AU - Jens Lundsgaard
AU - Colin Mikulski
AU - Zhixuan Yan
AU - Dhananjay Bhaskar
PY - 2026
DA - 2026/09/28
PB - arXiv
DO - 10.48550/arxiv.2609.34153
UR - https://arxiv.org/abs/2609.34153
AB - Proteins change shape as they function, yet most inverse folding models predict amino acid sequences from a single, fixed backbone. A central challenge in protein engineering is to design proteins that undergo specific motions, which requires accounting for how their structures change over time. This motivates inverse protein folding conditioned on protein motion. We introduce SPINET, which predicts sequences from molecular dynamics trajectories. It uses cellular sheaves to represent residue interactions within each frame and recurrent units to integrate information across frames, then predicts all amino acids in a single pass. We evaluate SPINET on mdCATH and ATLAS, where it outperforms all evaluated static and ensemble baselines in sequence recovery. On mdCATH, it achieves 56.7% top-1 recovery, compared with 44.5% for the strongest static baseline and 40.7% for the strongest ensemble baseline. We also evaluate whether the predicted sequences are compatible with conformations sampled along the target trajectory. On mdCATH, they achieve a median TM-score of 0.760, and structural recovery favors target conformations over unrelated decoys for 99.5% of test domains.
ER -