Xiaoqing Tan,Yongtao Tang,Tingting Yang,Guoliang Dai,Changqing Ye,Jianxin Meng,Fengyu Li
Photochromic sensors have the advantages of diverse isomers for multi-analysis, providing more sensing information and possessing more recognition units and more sensitivity to external stimulations, but they present enormous complexity with various stimulations as well. Deep learning (DL) algorithms contribute a huge advantage at analyzing nonlinear and multidimensional data, but they suffer from nontransparent inner networks, “black-boxes”. In this work, we employed the explainable DL approach to process and explicate photochromic sensing. Spirooxazine metallic complexes were adopted to prepare a multi-state analysis array for β-Lactams identification and quantitation. A dataset of 2520 unduplicated fluorescence intensity images was collected for convolutional neural network (CNN) operation. The method clearly discriminated six β-Lactams with 97.98% prediction accuracy and allowed rapid quantification with a concentration range from 1 to 100 mg/L. The photochromic sensing mechanism was verified via molecular simulation and class activation mapping, which explicated how the CNN model assesses the importance of photochromic sensor states and makes a discrimination decision. The explainable DL-assisted analysis method establishes an end-to-end strategy to ascertain and verify the complicated sensing mechanism for device optimization and even new scientific discovery.
Oxytetracycline (OTC), a persistent and poorly biodegradable antibiotic, is increasingly detected in aquatic environments, posing ecological and health risks. Herein, a defect-rich, Co-free MoS₂/PMS system enables ultrafast OTC removal, achieving >99% elimination within 5 min (k obs = 0.791 min⁻¹). The process remained highly efficient across pH 2–10 and in real waters containing natural organic matter and common inorganic ions, demonstrating strong robustness under complex water matrices. Radical-quenching and electron paramagnetic resonance (EPR) analyses confirmed that •OH was the dominant reactive oxygen species (ROS), accompanied by SO₄•⁻ and ¹O₂. X-ray photoelectron spectroscopy (XPS) evidenced Mo⁴⁺/Mo⁶⁺ redox cycling and the participation of surface sulfur species and lattice oxygen, which jointly facilitated PMS activation and ROS generation, thereby establishing a coupled radical-non-radical degradation mechanism. Liquid chromatography–mass spectrometry (LC–MS) identified sequential demethylation, deamination, and ring-opening intermediates, while QSAR-based ecotoxicity assessment indicated an overall decrease in predicted toxicity for most transformation products relative to OTC. Machine-learning analysis (XGBoost with SHAP interpretation) highlighted reaction time, PMS dose, and natural organic matter as the primary determinants of degradation efficiency. Overall, this work integrates a Co-free, light-independent MoS₂/PMS system, refined dual radical–non-radical mechanistic elucidation, comprehensive toxicity mitigation assessment and explainable machine learning, offering both mechanistic and engineering guidance for the sustainable treatment of antibiotic-contaminated wastewater.
Yanyi Yang,Xia Long,Jiaqing Dai,Xiaochi Liu,Duokai Zheng,Juexian Cao,Yong Hu
Interpretability is fundamental in the precise identification of single-molecule charge transport, and its absence in deep learning models is currently the major barrier to the usage of such powerful algorithms in the field. Here, we have pioneered a novel identification method employing fusion attention-based deep learning technologies. Central to our approach is the innovative neural network architecture, SingleFACNN, which integrates convolutional neural networks with a fusion of multihead self-attention and spatial attention mechanisms. Our findings demonstrate that SingleFACNN accurately classifies the three-type and four-type STM-BJ data sets, leveraging the convolutional layers’ robust feature extraction and the attention layers’ capacity to capture long-range interactions. Through comprehensive gradient-weighted class activation mapping and ablation studies, we identified and analyzed the critical features impacting classification outcomes with remarkable accuracy, thus enhancing the interpretability of our deep learning model. Furthermore, SingleFACNN’s application was extended to mixed samples with varying proportions, achieving commendable prediction performance at low computational cost. Our study underscores the potential of SingleFACNN in advancing the interpretability and credibility of deep learning applications in single-molecule charge transport, opening new avenues for single-molecule detection in complex systems.
危险品化学品经营许可证（带存储）