Problem
Research questions and friction points this paper is trying to address.
inverse protein folding
protein dynamics
molecular dynamics trajectories
protein engineering
conformational changes
SPINET: Sheaf Protein Inverse Folding Network
🤖 AI Summary
This study addresses the limitation of conventional inverse folding models that rely on static backbones, neglecting dynamic conformational changes essential for designing proteins with specific motile functions. To overcome this, we propose the first inverse folding paradigm conditioned on protein motion. Leveraging molecular dynamics trajectories, our approach employs cellular sheaves to characterize intra-frame residue interactions and integrates recurrent units to capture inter-frame temporal dependencies, enabling full-sequence prediction in a single forward pass. This work breaks through the constraints of static backbone representations. Evaluated on the mdCATH dataset, the proposed method achieves a Top-1 sequence recovery rate of 56.7%, substantially outperforming existing baselines. Furthermore, the predicted sequences demonstrate high compatibility with target conformations, yielding a median TM-score of 0.760.
📝 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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Inverse protein folding
Cellular sheaves
Molecular dynamics trajectories
Recurrent neural networks
Protein design
🔎 Similar Papers
No similar papers found.
