ShapeAligner: Direct Molecular Superposition from Clique-Based Atom Correspondence
7 affiliations
- 1Department of Chemistry
- 2University of Connecticut
- 3Electrical and Computer Engineering, Northwestern University, 2145 Sheridan Road, Evanston, IL, USA, 60208
- 4Department of Materials Science and Engineering and Institute of Materials Science
- 5Department of Pharmaceutical Sciences
- 6Department of Mechanical and Aerospace Engineering
- 7Texas Tech University
Abstract
Superposing one small molecule onto another is routine in ligand-based drug discovery. In atom-level correspondence methods, the pose the correspondence determines is only a starting point, and a subsequent search produces what is reported. We present ShapeAligner, a rigid aligner whose output pose is the superposition its maximal clique correspondence determines. Maximal cliques of an association graph over chemically typed heavy atoms each fix a reflection-free, equally weighted Kabsch transform in closed form. An analytic Gaussian shape-and-color volume overlap ranks the resulting poses. Nothing after the superposition moves the pose. Across 68,150 directed tasks from the LOBSTER cross-pair corpus, the top pose lands within 2.0 Å of the crystallographic pose in 30.98% of the tasks ShapeAligner covers, against 28.32% for the next-highest of six comparators. The median time is 0.214 s per task on one CPU core. Recovery depends strongly on the pair. It is highest when the larger ligand serves as the template, and the two ligands share at least 0.60 of the smaller ligand's volume. Below that threshold no method recovers more than 4% of the tasks it covers. The output is deterministic, and the correspondence it rests on is invariant to the input coordinate frame by construction. The implementation is released under the MIT license.
Related
- Direct Determination of Pyroelectric Coefficients in Molecular Crystals by X-ray Quantum Crystallography
- Direct separation of intra- and inter-molecular contributions in pulse dipolar EPR experiments
- Bottom-up design of the operating window for direct urea synthesis from first-principles molecular thermodynamics
- Direct visualization of heterogeneous molecular diffusion and interfacial barriers during condensate ripening
- Repetitive superposition performance of stress waves in control module of projectile under dynamic impact
BibTeX
@misc{ozuguzel2026shapealigner,
title = {{ShapeAligner: Direct Molecular Superposition from Clique-Based Atom Correspondence}},
author = {Umut Ozuguzel and Ugur Demir and Taradutt Pattnaik and Purvi Neema and S. Pamir Alpay and Bodhisattwa Chaudhuri and Adelia J. A. Aquino},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008368/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008368/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - ShapeAligner: Direct Molecular Superposition from Clique-Based Atom Correspondence
AU - Umut Ozuguzel
AU - Ugur Demir
AU - Taradutt Pattnaik
AU - Purvi Neema
AU - S. Pamir Alpay
AU - Bodhisattwa Chaudhuri
AU - Adelia J. A. Aquino
PY - 2026
DA - 2026/09/06
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008368/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008368/v1
AB - Superposing one small molecule onto another is routine in ligand-based drug discovery. In atom-level correspondence methods, the pose the correspondence determines is only a starting point, and a subsequent search produces what is reported. We present ShapeAligner, a rigid aligner whose output pose is the superposition its maximal clique correspondence determines. Maximal cliques of an association graph over chemically typed heavy atoms each fix a reflection-free, equally weighted Kabsch transform in closed form. An analytic Gaussian shape-and-color volume overlap ranks the resulting poses. Nothing after the superposition moves the pose. Across 68,150 directed tasks from the LOBSTER cross-pair corpus, the top pose lands within 2.0 Å of the crystallographic pose in 30.98% of the tasks ShapeAligner covers, against 28.32% for the next-highest of six comparators. The median time is 0.214 s per task on one CPU core. Recovery depends strongly on the pair. It is highest when the larger ligand serves as the template, and the two ligands share at least 0.60 of the smaller ligand's volume. Below that threshold no method recovers more than 4% of the tasks it covers. The output is deterministic, and the correspondence it rests on is invariant to the input coordinate frame by construction. The implementation is released under the MIT license.
ER -