OpenNovelty.org

SigmaDock: Untwisting Molecular Docking with Fragment-Based SE(3) Diffusion

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 6.0Download Report PDF
Molecular DockingGeometric Deep LearningGenerative ModelsSE(3)DiffusionAI for Science
Abstract:

Determining the binding pose of a ligand to a protein, known as molecular docking, is a fundamental task in drug discovery. Generative approaches promise faster, improved, and more diverse pose sampling than physics-based methods, but are often hindered by chemically implausible outputs, poor generalisability, and high computational cost. To address these challenges, we introduce a novel fragmentation scheme, leveraging inductive biases from structural chemistry, to decompose ligands into rigid-body fragments. Building on this decomposition, we present SigmaDock, an SE(3) Riemannian diffusion model that generates poses by learning to reassemble these rigid bodies within the binding pocket. By operating at the level of fragments in SE(3), SigmaDock exploits well-established geometric priors while avoiding overly complex diffusion processes and unstable training dynamics. Experimentally, we show SigmaDock achieves state-of-the-art performance, reaching Top-1 success rates (RMSD <2 & PB-valid) above 79.9% on the PoseBusters set, compared to 12.7-32.8% reported by recent deep learning approaches, whilst demonstrating consistent generalisation to unseen proteins. SigmaDock is the first deep learning approach to surpass classical physics-based docking under the PB train-test split, marking a significant leap forward in the reliability and feasibility of deep learning for molecular modelling.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Overall Novelty Assessment

The paper introduces SigmaDock, an SE(3) Riemannian diffusion model that generates ligand binding poses by reassembling rigid-body fragments within protein pockets. According to the taxonomy, this work resides in the 'Fragment-Based SE(3) Diffusion for Ligand Docking' leaf, which currently contains only this paper as its sole member. This placement indicates a relatively sparse research direction within the broader landscape of molecular docking methods, suggesting the approach occupies a distinct niche combining fragment decomposition with SE(3)-equivariant diffusion specifically for ligand pose prediction.

The taxonomy reveals that SigmaDock's immediate neighbors include SE(3) equivariant diffusion methods for peptide docking and flexible protein-ligand docking, as well as broader fragment-based diffusion approaches for structure-based drug design. While related branches explore autoregressive fragment assembly, dual equivariant architectures, and fragment-linker co-design, SigmaDock distinguishes itself by focusing on rigid-body fragment reassembly under SE(3) symmetry constraints. The taxonomy's scope notes clarify that peptide-specific methods and atom-level diffusion models belong elsewhere, positioning this work at the intersection of geometric deep learning and fragment-based molecular modeling.

Among the fourteen candidates examined during literature search, the contribution-level analysis found no clearly refuting prior work across the three main contributions: the fragment-based SE(3) diffusion framework itself, the FR3D fragmentation scheme with soft geometric constraints, and the SO(3)-equivariant architecture. The FR3D scheme was compared against four candidates without finding overlapping prior work, while the architectural contribution was evaluated against ten candidates with similar results. Given the limited search scope of fourteen papers from semantic search and citation expansion, these statistics suggest the specific combination of fragment-level SE(3) diffusion for docking appears relatively unexplored within the examined literature.

Based on the top-fourteen semantic matches and the taxonomy structure, the work appears to introduce a novel combination of fragment decomposition and SE(3)-equivariant diffusion for molecular docking. The analysis covers a focused subset of the literature rather than an exhaustive survey, and the absence of the paper's siblings in its taxonomy leaf suggests limited direct competition within this specific methodological niche. However, the broader field contains numerous fragment-based and equivariant diffusion approaches that may share conceptual overlap not captured by the limited search scope.

Taxonomy

Core-task Taxonomy Papers
38
3
Claimed Contributions
14
Contribution Candidate Papers Compared
0
Refutable Paper

Research Landscape Overview

Core task: molecular docking with fragment-based diffusion models. The field has evolved into a rich landscape where fragment-based strategies intersect with modern generative modeling. At the highest level, the taxonomy reveals several major branches: SE(3) equivariant diffusion methods that respect molecular symmetries for docking, fragment-based diffusion models tailored for structure-based drug design, diffusion approaches focused on molecular optimization and elaboration, general de novo design frameworks, non-diffusion fragment-based generative models, conformer generation and assembly techniques, classical and hybrid docking methods, fragment-based virtual screening pipelines, and overarching reviews integrating machine learning into docking workflows. Works such as DiffBindFR[30] and Dual Diffusion Binding[7] exemplify how SE(3) equivariance can be harnessed for physically plausible ligand placement, while methods like Autoregressive Fragment Diffusion[12] and Fragment Masked Diffusion[18] illustrate diverse strategies for assembling molecules from smaller building blocks. Within this ecosystem, a particularly active line of research explores how to balance geometric rigor with fragment-level modularity. Some approaches prioritize strict equivariance to ensure rotational and translational consistency during docking, whereas others emphasize flexible fragment assembly or optimization steps that refine initial placements. SigmaDock[0] sits squarely within the SE(3) equivariant diffusion branch for ligand docking, sharing conceptual ground with methods like DiffBindFR[30] that also leverage symmetry-aware generative processes. Compared to more general fragment-based frameworks such as Fragment Generative Framework[3] or Pocket Specific Fragment[1], SigmaDock[0] places stronger emphasis on maintaining geometric invariances throughout the diffusion process. This positioning highlights an ongoing tension in the field: whether to adopt highly structured, physics-informed generative models or to pursue more flexible, data-driven fragment assembly strategies that may sacrifice some geometric guarantees for greater design diversity.

Claimed Contributions

Fragment-based SE(3) diffusion model for molecular docking

The authors propose SIGMADOCK, a diffusion model that operates on rigid molecular fragments in SE(3) space rather than torsional angles. The model decomposes ligands into rigid-body fragments using structural chemistry priors and learns to reassemble them through Riemannian diffusion, avoiding the geometric entanglement and ambiguities inherent in torsional-space approaches.

0 retrieved papers
Fragment reduction scheme (FR3D) with soft geometric constraints

The authors introduce FR3D, a molecular fragmentation reduction algorithm that recursively merges adjacent fragments to minimize degrees of freedom while maintaining chemical validity. They also propose soft triangulation constraints that implicitly enforce bond length and angle preservation across fragments without restricting torsional freedom.

4 retrieved papers
SO(3)-equivariant architecture for fragment-based diffusion

The authors design an SO(3)-equivariant neural architecture that augments EquiformerV2 with hierarchical virtual nodes and edges, specialized featurizations for different structural roles, and smooth distance-based message decay. The architecture ensures invariance to local coordinate frame choices while maintaining equivariance properties.

10 retrieved papers

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Within the taxonomy built over the current TopK core-task papers, the original paper is assigned to a leaf with no direct siblings and no cousin branches under the same grandparent topic. In this retrieved landscape, it appears structurally isolated, which is one partial signal of novelty, but still constrained by search coverage and taxonomy granularity.

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

Fragment-based SE(3) diffusion model for molecular docking

The authors propose SIGMADOCK, a diffusion model that operates on rigid molecular fragments in SE(3) space rather than torsional angles. The model decomposes ligands into rigid-body fragments using structural chemistry priors and learns to reassemble them through Riemannian diffusion, avoiding the geometric entanglement and ambiguities inherent in torsional-space approaches.

Contribution

Fragment reduction scheme (FR3D) with soft geometric constraints

The authors introduce FR3D, a molecular fragmentation reduction algorithm that recursively merges adjacent fragments to minimize degrees of freedom while maintaining chemical validity. They also propose soft triangulation constraints that implicitly enforce bond length and angle preservation across fragments without restricting torsional freedom.

Contribution

SO(3)-equivariant architecture for fragment-based diffusion

The authors design an SO(3)-equivariant neural architecture that augments EquiformerV2 with hierarchical virtual nodes and edges, specialized featurizations for different structural roles, and smooth distance-based message decay. The architecture ensures invariance to local coordinate frame choices while maintaining equivariance properties.

Table of Contents

SigmaDock: Untwisting Molecular Docking with Fragment-Based SE(3) Diffusion | Novelty Validation