Related Experiment Video
Updated: Oct 8, 2025
Qualitative Identification of Carboxylic Acids, Boronic Acids, and Amines Using Cruciform Fluorophores
Published on: August 19, 2013
Explainable Deep Learning-Assisted Fluorescence Discrimination for Aminoglycoside Antibiotic Identification.
Xiaoqing Tan1, Yongpeng Liang1, Yingying Ye1
1College of Chemistry and Materials Science, Guangdong Provincial Key Laboratory of Functional Supramolecular Coordination Materials and Applications, Guangdong Engineering & Technology Research Centre of Graphene-like Materials and Products, Jinan University, Guangzhou 510632, China.
This study introduces an explainable deep learning (DL) method for analyzing complex sensor data, improving the detection of six aminoglycoside antibiotics (AGs) in various water types with high accuracy.
Area of Science:
- Analytical Chemistry
- Biomedical Engineering
- Data Science
Background:
- Current high-throughput sensing methods struggle with complex, multivariate biological and environmental data analysis.
- Deep learning (DL) excels at analyzing nonlinear, multidimensional data but often functions as a "black box."
Purpose of the Study:
- To develop an explainable DL-assisted visualized fluorometric array-based sensing method.
- To address the nontransparent inner workings of DL models in multianalyte identification.
Main Methods:
- Utilized a dataset of 8496 fluorometric images for training and validation.
- Investigated two DL algorithms (including CNN) and eight machine learning algorithms.
- Employed class activation mapping for model interpretability and a feedback mechanism for sensor array optimization.
Main Results:
- The convolutional neural network (CNN) achieved 100% prediction accuracy for six aminoglycoside antibiotics (AGs).
- A limit of detection of 1.34 ppm for AGs was established across diverse water samples (domestic, industrial, medical, consumption, aquaculture).
- Class activation mapping visualized the CNN's decision-making process, highlighting important sensor elements.
Conclusions:
- The explainable DL method provides an "end-to-end" strategy to overcome DL "black box" limitations.
- This approach facilitates hardware design optimization and creates facile indicators for environmental monitoring and disease diagnosis.
- The method promotes advancements in sensing technologies for scientific discovery and practical applications.
