PolyConf: Unlocking Polymer Conformation Generation through Hierarchical Generative Models

TL;DR

PolyConf tackles the gap in polymer conformation generation by introducing a frame-based, hierarchical approach that decouples repeating-unit conformations from their orientation transformations. Phase 1 uses a masked autoregressive diffusion process conditioned on a 2D polymer graph to sample repeating-unit conformations, while Phase 2 employs an SO(3) diffusion model to infer unit orientations and assemble the full polymer. The work also provides PolyBench, the first high-quality MD-derived benchmark for polymers, and demonstrates that PolyConf achieves state-of-the-art accuracy and efficiency, with strong scalability to larger polymer systems. This approach enables more realistic and rapid polymer conformations, facilitating improved polymer modeling and simulation in materials science and related fields.

Abstract

Polymer conformation generation is a critical task that enables atomic-level studies of diverse polymer materials. While significant advances have been made in designing conformation generation methods for small molecules and proteins, these methods struggle to generate polymer conformations due to their unique structural characteristics. Meanwhile, the scarcity of polymer conformation datasets further limits the progress, making this important area largely unexplored. In this work, we propose PolyConf, a pioneering tailored polymer conformation generation method that leverages hierarchical generative models to unlock new possibilities. Specifically, we decompose the polymer conformation into a series of local conformations (i.e., the conformations of its repeating units), generating these local conformations through an autoregressive model, and then generating their orientation transformations via a diffusion model to assemble them into the complete polymer conformation. Moreover, we develop the first benchmark with a high-quality polymer conformation dataset derived from molecular dynamics simulations to boost related research in this area. The comprehensive evaluation demonstrates that PolyConf consistently outperforms existing conformation generation methods, thus facilitating advancements in polymer modeling and simulation.

Figures (7)

• Figure 1: The comparison of small molecule, protein, and polymer. Here, the polymer chain comprises a series of repeating units, and the atomic force microscopy (AFM) image roiter2005afm is used for direct 3D visualization of various polymer chains.
• Figure 2: The overview of PolyConf: A hierarchical framework for polymer conformation generation, employing a masked autoregressive (MAR) model with a diffusion procedure to sample the conformation of each repeating unit within the polymer in random order, followed by a SO(3) diffusion model to assemble these repeating unit conformations into the complete polymer conformation. Here, for the sake of visualization simplicity, only a small fragment of the whole polymer conformation with five repeating units is presented in this figure.
• Figure 3: The illustration of frame-based polymer representation. In particular, the 1D polymer SMILES string represents the monomer's SMILES string with two "*" symbols marking polymerization sites. The 3D polymer conformation comprises a series of repeating unit conformations with identical SMILES strings but distinct 3D structures, overlapping at key atoms (e.g., atom-1 aligns with atom-3 of the previous repeating unit). The orientation transformation, derived from the key atoms within the corresponding frame, is denoted as $\mathcal{O} = (\bm R, \bm t)$ where the rotation $\bm R \in \mathbb{R}^{3 \times 3}$ is calculated through the GramSchmidt operation leon2013gram on vectors $\bm v_1$ and $\bm v_2$, and the translation $\bm t \in \mathbb{R}^{3}$ corresponds to the 3D coordinate of atom-3.
• Figure 4: The illustration of the masked autoregressive modeling in the second phase, where grey blocks represent the corresponding embeddings of masked repeating units.
• Figure 5: The efficiency comparison (average time) of various methods on the PolyBench benchmark, where we compare the average time of generating polymer conformations.