Circular Stereo Algorithm and Fingerprint for Chiral Resonance Invariant Molecular Representation
Abstract
Algorithmic molecular fingerprints remain attractive for high-throughput chemoinformatics because they are fast, discrete, and easy to integrate into screening, similarity search, and machine-learning workflows. However, widely used circular fingerprints such as ECFP rely on conventional molecular graph representations that capture stereochemistry incompletely and may distinguish chemically equivalent resonance forms through localized atom charges and bond-order-dependent encodings. Here, we introduce the Circular Stereo Algorithm (CSA) and the derived Circular Stereo Fingerprint (CSFP), a stereochemically informed, resonance-invariant extension of circular graph refinement built on the StereoMolGraph representation.[Papusha and Leonhard, J. Chem. Inf. Model, 2026, 66, 7, 3830] In this framework, stereochemistry is encoded through local stereodescriptors grounded in group theory, allowing atom-centered and bond-centered stereochemical information to be incorporated directly while avoiding artificial distinctions between resonance structures. We show that CSA yields locally consistent atom invariants for stereoisomers, preserves equivalence across resonance forms, and discriminates challenging stereochemical cases, including delocalized radicals and metal complexes. Compared with ECFP on stereoisomer datasets, CSFP produces less over-discriminating atom identifiers and a more reliable mapping between subgraphs and fingerprints. Evaluation on 1.4 million ChEMBL small molecules shows a low color-refinement-hash collision rate of 0.1%, with remaining ambiguities resolvable by exact StereoMol-Graph isomorphism. We showcase how its mathematical properties make it suitable for substructure search on large-scale datasets.
Authors 2
-
Maxim Papusha Aachen
Affiliation as printed
Institute of Technical Thermodynamics
RWTH Aachen University
-
Kai Leonhard Aachen
Affiliation as printed
Institute of Technical Thermodynamics
RWTH Aachen University
Cited by 0 stored of 0
No patents citing this paper on Lens.org (checked 2026-10-06).