Transformer, a sequence-to-sequence deep learning model, is capable of predicting the reaction intermediates of radical cascade cyclization. This study provides a novel approach to help chemists discover the mechanisms of organic reactions.

Fluorescent drug molecules play a pivotal role in biomedical research and precision medicine. Their intrinsic fluorescence enables real-time tracking of drug distribution, target engagement, and metabolic pathways, while avoiding interference from external labeling. However, traditional fluorescent drug discovery relies heavily on trial-and-error approaches, which are inefficient and resource-intensive. To address this, we developed DyeLeS (Dye-Likeness Scoring), a web platform designed to rapidly evaluate molecular fluorescence potential and predict key photophysical properties such as Stokes shift and quantum yield. DyeLeS utilizes a curated dataset of fluorescent and non-fluorescent compounds, applies a Naive Bayes-inspired algorithm for fluorescence classification (AUC = 0.995), and employs a LightGBM model for quantitative prediction of fluorescence properties, achieving an R² of 0.88 in absorption wavelength (λ_abs) prediction. Leveraging DyeLeS, this study constructed FluoBioDB, the first publicly available library of fluorescent bioactive compounds, encompassing 32 865 structurally diverse molecules, including kinase inhibitors and GPCR modulators. Case analyses indicate that FluoBioDB compounds typically possess polycyclic conjugated frameworks, donor–acceptor (D–A) structures, and rigid planar cores, endowing them with strong potential for applications in bioimaging, targeted therapy, and theranostics. This work presents a robust computational framework and a valuable molecular resource to facilitate the rapid discovery and optimization of fluorescent drug candidates. All source codes and datasets are available at https://github.com/MolAstra/DyeLeS, and the web server can be accessed at https://dyeles.molastra.com.

The eEF2K, a member of the α-kinase family, plays a crucial role in cellular differentiation and the stability of the nervous system. The development of eEF2K inhibitors has proven to be significantly important in the treatment of diseases such as cancer and Alzheimer’s. With the advancement of big data in pharmaceuticals and the evolution of molecular generation technologies, leveraging artificial intelligence to expedite research on eEF2K inhibitors shows great potential. Based on the recently published structure of eEF2K and known inhibitor molecular structures, a generative model was used to create 1094 candidate inhibitor molecules. Analysis indicates that the model-generated molecules can comprehend the principles of molecular docking. Moreover, some of these molecules can modify the original molecular frameworks. A molecular screening strategy was devised, leading to the identification of five promising eEF2K inhibitor lead compounds. These five compound molecules demonstrated excellent thermodynamic performance when docked with eEF2K, with Vina scores of −12.12, −16.67, −15.07, −15.99, and −10.55 kcal/mol, respectively, showing a 24.27% improvement over known active inhibitor molecules. Additionally, they exhibited favorable drug-likeness. This study used deep generative models to develop eEF2K inhibitors, enabling the treatment of cancer and neurological disorders.

With the rapid development of radical initiation technologies such as photocatalysis and electrocatalysis, radical reactions have become an increasingly attractive approach for constructing target molecules. However, designing efficient synthetic routes using radical reactions remains a significant challenge due to the inherent complexity and instability of radical intermediates. While computer-aided synthesis planning (CASP) has advanced retrosynthetic analysis for polar reactions, radical reactions have been largely overlooked in AI-driven approaches. In this study, we introduce RadicalRetro, the first deep learning-based retrosynthesis model specifically tailored for radical reactions. Our work is distinguished by three key contributions: (1) RadicalDB: A novel, manually curated database of 21.6 K radical reactions, focusing on high-impact literature and mechanistic clarity, addressing the critical gap in dedicated radical reaction datasets. (2) Model Innovation: By pretraining Chemformer on ZINC-15 and USPTO datasets followed by fine-tuning with RadicalDB, RadicalRetro achieves a Top-1 accuracy of 69.3% in radical retrosynthesis, surpassing the state-of-the-art models LocalRetro and Mol-Transformer by 23.0% and 25.4%, respectively. (3) Interpretability and Practical Utility: Attention weight analysis and case studies demonstrate that RadicalRetro effectively captures radical reaction patterns (e.g., cascade cyclizations and photocatalytic steps) and proposes synthetically viable routes, such as streamlined pathways for Tamoxifen precursors and glycoside derivatives. RadicalRetro’s performance highlights its potential to transform radical-based synthetic planning, offering chemists a robust tool to leverage the unique advantages of radical chemistry in drug synthesis.

Photocatalysis is becoming increasingly important in modern chemistry for efficient multicomponent one-pot synthesis. However, predicting the results of photocatalytic reactions using artificial intelligence remains challenging, mostly due to the insufficient number of photocatalytic reactions and the incomplete information on reaction conditions in existing reaction databases. In this study, we curated the Photocatalysis Database (PhotoCatDB), which consists of 6,523 photocatalytic reactions (of which 6,175 are multicomponent) containing reaction condition information such as photocatalysts, bases or acids, additives, and solvents. Before adding reaction conditions to the training data, the attention-based deep learning model PhotoCat pre-trained on USPTO and fine-tuned on PhotoCatDB had a Top-1 accuracy of 78.16%, which was 77.70% higher than the same model trained only on the USPTO database and 14.53% higher than the model fine-tuned by the photocatalytic reactions from Reaxys. After adding reaction conditions to the training data, the Top-1 accuracy of PhotoCat was further increased to 82.25%. In addition, the interpretability of the model was reflected in its attention weights, which can infer the model’s understanding of photocatalytic chemistry. Furthermore, five previously unreported photocatalytic reactions predicted by PhotoCat were successfully validated by wet-lab experiments, demonstrating the potential of the model in identifying and verifying novel photocatalysis reactions of real-world significance.

With the rapid development of radical initiation technologies such as photocatalysis and electrocatalysis, radical reactions have become an increasingly attractive approach for constructing target molecules. However, determining how to efficiently synthesize a specific target molecule using radical reactions remains a challenging task, requiring years of education, training, and extensive trial-and-error by synthetic chemists. The advancement of computer-aided synthesis planning (CASP) has opened new possibilities for accelerating the optimization of radical reaction pathways through artificial intelligence. In this study, we collected 21.6K entries of radical reaction data to build the Radical Reaction Database (RadicalDB). After pre-training with ZINC-15 and USPTO database, we fine-tuned the Chemformer model using RadicalDB, resulting in the development of RadicalRetro, a retrosynthetic prediction model specialized in radical reactions. RadicalRetro achieved a top-1 prediction accuracy of 69.3% for radical retrosynthesis, outperforming similar models such as LocalRetro and Mol-Transformer by 23.0% and 25.4%, respectively. Case studies and attention weight analyses confirm that RadicalRetro effectively captures the unique characteristics of radical reactions.