molecular fingerprinting

General-purpose molecular representations lack solvent-specific physical grounding, while hazardous solvent usage constitutes a major climate risk in chemical engineering. Green solvent substitution research faces dual challenges: inadequate solvent representation and scarce labeled data. Method: We propose SoDaDE—a lightweight Transformer-based embedding method trained on small-scale solvent property data—marking the first application of compact Transformers to solvent representation learning. It generates continuous, physics-interpretable fingerprint vectors via self-supervised learning, enabling fine-grained, domain-adapted representation even under data scarcity. Contribution/Results: SoDaDE overcomes the physical semantic deficiency inherent in conventional molecular fingerprints. Evaluated on a state-of-the-art solvent dataset, it significantly outperforms traditional representations in downstream tasks such as reaction yield prediction, demonstrating its efficacy and feasibility for high-performance chemical representation learning in low-data regimes.

This study addresses three critical challenges in chemical process design: low accuracy in molecular property prediction, inefficient discovery of novel molecules, and the absence of molecular–process co-design. To tackle these, we propose a hybrid architecture integrating physics- and chemistry-informed graph neural networks (GNNs) with Transformer modules, enabling high-accuracy, interpretable prediction of thermodynamic and transport properties for both pure components and mixtures. Building upon this, we establish a closed-loop framework unifying molecular generation, property prediction, and process optimization—facilitating targeted exploration of chemical space and cross-scale co-design. Furthermore, we introduce the first unified benchmark spanning molecular modeling, process simulation, and industrial validation, substantially enhancing model generalizability and engineering applicability. The resulting paradigm provides a scalable, experimentally verifiable methodology for the co-development of novel functional molecules and low-carbon chemical processes.

Traditional reaction rules rely on manually encoded heuristics, limiting their coverage of the long-tail chemical space and adaptability to novel transformations. This work proposes an automated, self-extending framework for reaction rule generation based on a multi-agent large language model, enabling human-intervention-free rule evolution through a closed-loop validation mechanism. By integrating a lightweight fingerprint classifier with a symbolic system, the method achieves 97.7% classification accuracy on 665,901 unseen patent reactions, expanding standard reaction categories from 68 to 14,073—substantially outperforming leading commercial classifiers. Furthermore, it supports immediate generalization to out-of-distribution reactions, thereby establishing a sustainably evolving reactivity database.

Modeling the relationship between molecular structure and odor perception in fragrance compounds remains challenging due to structural complexity and poor model interpretability. Method: We propose the first end-to-end graph generation framework integrating molecular graph generation, structural validity verification, fragrance propensity filtering, and multi-level odor attribute prediction (20 odor classes). Our approach couples graph neural networks (GNNs) with conditional variational autoencoders (CVAEs), jointly optimizing graph generation and high-accuracy multi-task odor prediction. We incorporate SHAP for post-hoc interpretation, revealing causal links between physicochemical features and odor attributes, and embed a molecular sanitization rule engine to enforce chemical validity. Contribution/Results: The framework achieves state-of-the-art odor prediction performance (ROC AUC = 0.97) and enhanced interpretability. We publicly release code and pretrained models, establishing a new computational paradigm for fragrance science and olfactory mechanism research.

Reaction prediction requires modeling the intrinsic stochasticity of electron redistribution; however, conventional maximum-likelihood generative models (e.g., Transformers) output only the most probable product, failing to capture rare yet high-value reaction pathways. To address this, we propose a collaborative multi-expert framework: (1) a novel “Reaction Expert Team” architecture that emulates chemists’ divergent thinking by assigning heterogeneous experts to model distinct electron-transfer pathways; (2) a ranking expert for credibility-weighted ensemble integration; and (3) synergistic components including Transformer-specialized expert clusters, reaction-pattern decoupled training, multi-objective consistency distillation, and ranking-aware enhancement. Evaluated on USPTO and MIT benchmarks, our method improves Top-5 diversity by 42% and boosts recall of rare high-value pathways by 3.8× over state-of-the-art methods.

This work reveals a “Clever Hans” effect in chemical machine learning models for activity prediction—models inadvertently rely on chemist-specific stylistic cues (e.g., synthetic preferences, author identity) rather than genuine structure–activity causal relationships. Method: We systematically demonstrate that molecular fingerprints enable high-accuracy author attribution (top-5 accuracy: 60%), and show that activity predictors trained solely on author-probability vectors match the performance of structure-based baselines—confirming severe causal leakage. To mitigate bias, we propose author-disjoint data splits, scaffold-aware partitioning, and explicit author-probability modeling. Contribution/Results: Our findings motivate the “de-intentionalization” of benchmark datasets in cheminformatics—removing confounding author-level signals—and establish theoretical foundations and methodological guidelines for building causally robust models in drug discovery. These practices enhance model generalizability and scientific interpretability by enforcing structural causality over spurious correlations.

This study addresses the limited functional specificity of traditional chemical functional groups in interpreting biological activity. We propose an unsupervised substructure discovery method grounded in the Minimum Message Length (MML) principle—marking the first integration of data compression theory with functional group identification. The algorithm automatically extracts compressible, functionally specific substructures from a corpus of three million biologically relevant molecules, enabling dataset-adaptive functional motif mining. Based on these substructures, we construct a novel molecular fingerprint that significantly outperforms MACCS and Morgan fingerprints across 24 biological activity prediction tasks, enhancing ridge regression model performance. Our core contribution is establishing a compression-driven paradigm for functional substructure discovery, unifying interpretability, functional specificity, and predictive accuracy in molecular representation learning.

This work addresses the inconsistency in predictions and explanations arising from multiple graph representations of the same molecule in molecular graph machine learning, which violates chemical identity. To resolve this, the authors propose InChIfied Invariants—strictly invariant features constructed at node, edge, and graph levels based on the International Chemical Identifier (InChI)—that inherently guarantee identical representations, predictions, and attributions for chemically equivalent graphs. Evaluation on the large-scale PubChem Substances dataset demonstrates that the method achieves consistent representations for 99.62% of chemically equivalent graph pairs, a dramatic improvement over the 0.35% consistency achieved by conventional Daylight invariants. Furthermore, it maintains competitive predictive performance on MoleculeNet benchmark tasks while significantly enhancing model interpretability and chemical plausibility.

This study addresses a critical yet overlooked issue in materials discovery: machine learning models may exploit non-chemical confounding factors—such as authorship, journal, and publication year (collectively termed “literature fingerprints”)—to achieve deceptively high predictive performance without genuine understanding of material chemistry. The work presents the first systematic investigation into how such metadata can induce spurious correlations, and introduces a novel evaluation paradigm to disentangle true chemical insight from artifacts of bibliographic bias. Through a suite of rigorous analyses—including modeling with standard chemical descriptors, metadata-only prediction, secondary modeling, grouped or temporal data splits, and metadata ablation—the authors demonstrate across five diverse materials tasks that literature fingerprints alone can yield performance nearly matching conventional approaches, thereby exposing significant hidden biases in widely used datasets.

Existing benchmarks for chemical reasoning evaluation focus solely on final answers, making it difficult to detect logical errors in intermediate reasoning steps. To address this limitation, this work proposes ChemCoTBench-V2, a novel benchmark that employs expert-designed structured templates to guide models in generating verifiable intermediate reasoning states. By integrating deterministic chemical rules, reference trajectory alignment, and oracle-verifiable state constraints, the framework enables low-cost, auditable process-level evaluation without requiring human or LLM-based adjudication. This approach is the first to support state-constraint verification and precise error localization in open-ended tasks, revealing a significant discrepancy between answer correctness and reasoning consistency across mainstream large language models. It further facilitates fine-grained model comparison and identification of the first erroneous step in reasoning trajectories.