(Tracing the evolution of artificial intelligence toward stereochemistry-aware molecular intelligence)
Message-passing networks for molecules with tetrahedral chirality
Researchers demonstrated that conventional graph-based neural architectures can be insensitive to stereochemistry and developed chirality-aware aggregation functions to learn properties associated with tetrahedral chirality. The work marked an early explicit attempt to incorporate molecular stereochemistry into neural molecular representations.
Learning 3D representations of molecular chirality
The development of ChIRo (Chiral InterRoto-Invariant Neural Network) represented a further step toward learning stereochemical information directly from three-dimensional molecular structures. The approach was designed to distinguish stereoisomers and demonstrated chiral-sensitive performance across several molecular tasks.
Research increasingly recognizes that how a molecule is represented computationally can strongly influence what an AI model can learn. A major review categorized molecular representations into string, connection-table, feature-based, and computer-learned approaches, providing an important framework for understanding the representation challenge in chemical AI.
ChiENN introduced a chirality-sensitive graph neural network layer designed to distinguish enantiomeric structures. The work demonstrated that explicitly incorporating chirality into graph-based architectures can improve performance in chiral-sensitive molecular property-prediction tasks.
Research on Transformer architectures trained on chemical string representations demonstrated that models can struggle to recognize molecular chirality. The findings highlighted an important limitation of relying on conventional molecular strings when stereochemical information is essential.
AI research moves beyond simply representing or recognizing chirality toward stereochemistry-aware molecular generation and explicit molecular chirality classification. The emergence of these approaches signals a shift toward incorporating stereochemical fidelity into generative and predictive AI systems.
The scope broadens from molecular representation and classification to molecular and reaction graphs, stereoselectivity prediction, asymmetric catalysis, and stereochemical assignment of natural products. These developments indicate an emerging move toward AI systems capable of handling stereochemical information across a wider range of chemical problems.
From molecular representation to Chiral Intelligence, the trajectory is clear: AI is moving from learning molecules in two dimensions toward understanding their three-dimensional stereochemical identity.
(Key contributions that have shaped the development of molecular representation, chirality recognition, stereochemistry-aware AI, and related applications presented in a consistent citation format with a brief “Why it matters” explanation and DOI where available)
Lagnajit Pattanaik, Octavian-Eugen Ganea, Ian Coley, Klavs F. Jensen, William H. Green, and Connor W. Coley.
Message Passing Networks for Molecules with Tetrahedral Chirality. Machine Learning for Molecules Workshop, NeurIPS 2020.
Why it matters: This work explicitly addressed the inability of conventional molecular graph architectures to distinguish stereochemical structures and introduced chirality-aware message-passing functions for tetrahedral chirality.
Keir Adams, Lagnajit Pattanaik, and Connor W. Coley.
Learning 3D Representations of Molecular Chirality with Invariance to Bond Rotations. International Conference on Learning Representations (ICLR). 2022.
Why it matters: The ChIRo model introduced a 3D, chirality-sensitive representation capable of learning stereochemical information from molecular conformations and was evaluated on tasks including R/S classification, optical rotation, and enantioselective docking.
Note: This work appeared as an ICLR 2022 conference paper; its preprint dates to 2021.
Daniel S. Wigh, Jonathan M. Goodman, and Alexei A. Lapkin.
A review of molecular representation in the age of machine learning. WIREs Computational Molecular Science. 2022;12(5):e1603.
Why it matters: This influential review places molecular representation at the heart of machine learning for chemistry and provides an important conceptual foundation for understanding why stereochemical information must be faithfully represented for AI.
Piotr Gaiński, Michał Koziarski, Jacek Tabor, and Marek Śmieja.
ChiENN: Embracing Molecular Chirality with Graph Neural Networks. Machine Learning and Knowledge Discovery in Databases: Research Track. Lecture Notes in Computer Science. 2023;14171:36–52.
Why it matters: ChiENN introduced a chirality-sensitive graph neural network layer that explicitly distinguishes enantiomeric structures and demonstrated improved performance on chiral-sensitive molecular property-prediction tasks.
Yasuhiro Yoshikai, Tadahaya Mizuno, Shumpei Nemoto, and Hiroyuki Kusuhara.
Difficulty in chirality recognition for Transformer architectures learning chemical structures from string representations. Nature Communications. 2024;15:1197.
Why it matters: This study provided direct evidence that Transformer models learning chemical structures from string representations can struggle with chirality recognition, exposing a significant chiral blind spot in AI.
Gary Tom, Edwin Yu, Naruki Yoshikawa, Kjell Jorner, and Alán Aspuru-Guzik.
Stereochemistry-aware string-based molecular generation. PNAS Nexus. 2025;4(11):pgaf329.
Why it matters: This study moves beyond chirality recognition toward stereochemistry-aware molecular generation, providing benchmarks for assessing whether generative AI preserves and exploits stereochemical information.
Yichuan Peng, Gufeng Yu, Runhan Shi, Letian Chen, Xi Wang, Wenjie Du, Xiaohong Huo, and Yang Yang.
ChiralCat: Molecular chirality classification with enhanced spatial representation using learnable queries. Artificial Intelligence Chemistry. 2025;3(2):100091.
Why it matters: ChiralCat advances AI-based molecular chirality classification through enhanced spatial representation and learnable queries, providing a dedicated approach to capturing stereochemical differences that conventional molecular representations may miss.
Maxim Papusha and Kai Leonhard.
StereoMolGraph: Stereochemistry-Aware Molecular and Reaction Graphs. Journal of Chemical Information and Modeling. 2026;66(7):3830–3839.
Why it matters: StereoMolGraph extends stereochemistry-aware representation to molecular and reaction graphs, addressing stereochemical information in both molecular structures and reactions.
Li Cheng, Pan-Lin Shao, Shaokang Zhao, Bo Zhang, and Guichuan Xing.
Capturing stereochemical information with AI: Driving stereoselectivity prediction and rational design in asymmetric catalysis. Chinese Chemical Letters. 2026;113075.
Why it matters: This review broadens the scope of stereochemistry-aware AI to stereoselectivity prediction, asymmetric catalysis, catalyst discovery, and AI-assisted catalyst design.
Markus Orsi and Jean-Louis Reymond.
Assigning the stereochemistry of natural products by machine learning. Journal of Cheminformatics. 2026;18(1):76.
Why it matters: This work demonstrates the application of machine learning to automated stereochemical assignment in natural products, extending stereochemistry-aware AI into structural elucidation.
