ProtScape: A molecular structure and energy-aware representation for protein conformation generation
Authors: Siddharth Viswanath, Xingzhi Sun, Lucas Lee, Danqi Liao, Hiren Madhu, David R. Johnson, João Felipe Rocha, Egbert Castro, +4 more
Organizations: Department of Computer Science, Yale University · Department of Genetics, Yale University · Program in Computing, Boise State University · Computational Biology and Bioinformatics Program, Yale University · Department of Mathematics, Boise State University
Molecular dynamics (MD) simulations are a principled but computationally expensive approach for studying protein conformational variability, making it challenging to generate large ensembles of structures or characterize transitions between metastable conformations. AI methods for upsampling MD simulations have been developed recently but struggle due to difficulties of sampling complex, high-dimensional molecular distributions. One avenue these methods overlook is to learn a latent space where such sampling becomes easier. To address this, we introduce ProtScape, a generative geometric deep learning framework that learns structure and energy-aware representations of protein conformational landscapes from MD simulations. ProtScape represents protein conformations using an equivariant graph neural network and a multiscale deep wavelet transform that captures local geometric interactions and nonlocal collective motions. This latent representation is dual-organized: 1) by structure captured by wavelet transform layers, via reconstruction error; 2) by energy using a Laplacian energy smoothness penalty. This structured latent space enables meaningful generation and exploration of protein conformations. ProtScape supports multiple generative modes: 1) ensemble generation, which upsamples ensembles given limited MD trajectories flow matching from noise to the organized manifold of conformations, 2) minimum-energy path generation between two high-energy conformations, guided by energy using a nudged elastic band, and 3) energy-descent, which generates trajectories toward lower energies from a high-energy conformation using gradient descent, thereby hypothesizing folding or other stabilizing trajectories. Organizing a latent space by structure and energy lets one representation support ensemble sampling, minimum-energy path finding and energy-guided descent on the same conformational landscape.
Figures & tables
Figure 1: Overview of the ProtScape framework. The model encodes atomistic protein features into a low-dimensional latent manifold using an SE(3)-equivariant network with deep graph wavelet transforms. The latent space is regularized by potential energy to ensure physical validity. This representation supports three complementary tasks: Ensemble Generation , utilizing latent diffusion to sample diverse equilibrium states, Minimum Energy Path Interpolation , employing Latent NEB to optimize minimum energy paths between conformations, and Energy-Guided Generation , performing energy-guided latent dynamics to iteratively refine conformations toward lower-energy and structurally stable states.
Figure 2: PHATE embeddings of conformational ensembles colored by potential energy. Rows correspond to three proteins (6p5hA, 7jflC, 7lp1A); columns correspond to ProtScape, MDGen, and AlphaFlow. Points are colored by potential energy. ProtScape creates a continuous landscape organized by energy with multi-modal distributions reflecting conformation variation; MDGen yields branched, discontinuous structure; AlphaFlow collapses to 1-dimensional trajectories.
Protein
Method
RMWD ↓
Trans. ↓
Var. ↓
Pool PCA-W 2↓
Ref PCA-W 2↓
6H86 (Chain A)
AlphaFlow
4.2487 ± 0.017
2.4578 ± 0.005
3.4657 ± 0.021
34.9194 ± 0.195
34.6883 ± 0.195
MDGen
1.0066 ± 0.008
0.8097 ± 0.008
0.5978 ± 0.014
3.8805 ± 0.209
3.7751 ± 0.207
\Block [fill=gray!15]1- ProtScape (DDPM)
0.3441 ± 0.016
0.2070 ± 0.013
0.2749 ± 0.010
1.4490 ± 0.215
1.4576 ± 0.215
\Block [fill=gray!15]1- ProtScape (Flow Matching)
0.4244 ± 0.104
0.2555 ± 0.063
0.3386 ± 0.083
1.8907 ± 1.473
1.8973 ± 1.475
6P5H (Chain A)
AlphaFlow
1.0504 ± 0.061
0.7488 ± 0.077
0.7350 ± 0.010
8.0587 ± 0.839
8.0806 ± 0.830
MDGen
0.6200 ± 0.009
0.4258 ± 0.013
0.4504 ± 0.015
2.2015 ± 0.131
2.0740 ± 0.140
Table 1: Distributional ensemble quality metrics for proteins in our dataset. Lower is better for all reported metrics. Best result in each category is bolded. Values are reported as mean ± std.
Protein
Method
Rama. Outliers (%) ↓
Clash Score ↓
Rotamer Outliers ↓
MolProbity Score ↓
Δ RMSD (Å) ↓
6H86 (Chain A)
MDGen
53.56 ± 11.87
397.89 ± 46.74
5.03 ± 0.11
16.30 ± 4.60
2.10 ± 0.56
\Block [fill=gray!15]1- ProtScape
1.70 ± 1.65
99.27 ± 99.80
0.00 ± 0.00
2.44 ± 0.82
0.20 ± 0.05
6P5H (Chain A)
MDGen
43.26 ± 7.17
371.61 ± 38.88
4.85 ± 0.11
11.54 ± 3.16
1.70 ± 0.29
\Block [fill=gray!15]1- ProtScape
3.38 ± 2.16
124.37 ± 108.04
0.00 ± 0.00
2.73 ± 0.67
0.113 ± 0.028
6JV8 (Chain A)
MDGen
54.95 ± 8.15
371.04 ± 53.68
4.90 ± 0.14
12.10 ± 4.01
2.00 ± 0.38
\Block [fill=gray!15]1- ProtScape
5.57 ± 2.85
154.15 ± 133.61
0.00 ± 0.00
2.81 ± 0.97
0.114 ± 0.037
Table 2: Minimum Energy Path using NEB. Structural validity is measured via Ramachandran outliers, steric clash score, rotamer outliers, and MolProbity score. We also report Δ RMSD (in Å; lower is better). Values are reported as mean ± std.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Figure 3: Isolated conformational potential energy profiles across the path frames for the five protein systems. The energy profiles reveal the pure rotational activation barriers, characterized by the steric transition state regions peaking near the path midpoints (Frames 40–60)
Protein dynamics underlie many biological functions, yet remain difficult to characterize due to the high computational cost of molecular dynamics simulations and the scarcity of dynamic structural data. This survey reviews recent advances in artificial intelligence for protein dynamics from three perspectives: learning from structural ensembles and trajectories, learning from physical energy signals, and learning to accelerate molecular simulations. We summarize representative methods for conformation ensemble generation, trajectory generation, Boltzmann generators, physics-aware adaptation, machine learning potentials, coarse-grained modeling, and collective variable discovery. We further discuss available datasets and key open challenges, such as scalability, thermodynamic consistency, kinetic fidelity, and integration with experimental constraints.
Haocheng Tang, Liang Shi, Ya-Shi Zhang +3
Mila – Quebec AI Institute · University of Pittsburgh · Work conducted during an internship at Mila +4
De novo protein generation has transformative potential in therapeutic design, enzyme engineering, and synthetic biology. While diffusion-based and flow matching approaches have achieved progress, they typically operate at single resolution and lack mechanisms for incorporating functional constraints. We introduce ProHiFlo, a hierarchical flow matching framework with three innovations: (1) coarse-to-fine generation that models backbone geometry before refining to all-atom coordinates, reducing computational cost while maintaining accuracy; (2) functional guidance leveraging pretrained predictors to steer generation toward desired properties without retraining; (3) adaptive SE(3)-equivariant architecture for efficient multi-scale processing. Experiments on unconditional generation, motif scaffolding, and functional design demonstrate state-ofthe-art performance while requiring 4 fewer sampling steps. On enzyme active site scaffolding, ProHiFlo achieves 58.9% success rate compared to 41.2% for RFDiffusion.
Chuanzhen Wang, Meade Cleti, Pete Jano
3Tongji University · 1Arizona State University · University of Wisconsin-Madison
Generative emulators of protein dynamics produce plausible trajectories at a fraction of the cost of molecular dynamics, but they inherit their training distribution and tend to revisit known states rather than reach rare ones under long-horizon extrapolation. Inspired by classical enhanced sampling, we introduce an implicit, history-dependent bias in the generative space of a pretrained emulator. Specifically, a history-aware score estimator augments the frozen emulator with a distance-weighted bias that steers reverse-time sampling away from previously generated structures, regularized by an environment-support term. To preserve structural validity at long horizons, a score-based refinement step re-projects drifted samples onto the data manifold using the frozen emulator. Our experiments demonstrate that the method (i) raises diversity by 35% on DynamicPDB-80; (ii) on 12 zero-shot Fast-Folding proteins, the learned bias alone reaches the unbiased emulator's coverage up to ∼15× faster, and pairing it with refinement reaches the coverage up to ∼37× faster while covering ∼3× as many low-energy states. Code will be released soon.
Kaihui Cheng, Zhiqiang Cai, Wenkai Xiang +4
1Fudan University · 2Shanghai Academy of AI for Science · 3Shanghai Innovation Institute