Related Experiment Video
Updated: Nov 28, 2025
08:01
Author Spotlight: Unraveling Bacterial Responses to Antibiotics and Immune System in Tissues
Published on: March 1, 2024
1.2K
Pre-Trained Deep Convolutional Neural Network for Clostridioides Difficile Bacteria Cytotoxicity Classification Based
Andrzej Brodzicki1, Joanna Jaworek-Korjakowska1, Pawel Kleczek1
1Department of Automatic Control and Robotics, AGH University of Science and Technology, 30-059 Kraków, Poland.
Sensors (Basel, Switzerland)
|December 1, 2020
Summary
Novel deep learning methods can now rapidly screen for new Clostridioides difficile infection (CDI) therapies by analyzing cell images. This computer-aided approach overcomes the limitations of manual screening for drug discovery, addressing an unmet clinical need.
Area of Science:
- Computational biology
- Infectious diseases
- Machine learning applications in drug discovery
Background:
- Clostridioides difficile infection (CDI) is a growing global health concern, with symptoms ranging from mild diarrhea to severe colitis.
- Antibiotics, the current standard treatment for CDI, paradoxically increase the risk of developing and recurring the infection.
- There is a critical need for novel therapies that can effectively treat CDI and prevent its recurrence.
Purpose of the Study:
- To develop and evaluate a novel computer-aided method for screening potential drug leads against CDI.
- To identify characteristic morphological changes in human fibroblast cells exposed to C. difficile toxins using computer vision and deep learning.
- To overcome the limitations of manual image analysis in drug discovery screening, such as being slow, tedious, and prone to error.
Main Methods:
- Utilized classical image processing algorithms for pre-processing fluorescence images of human fibroblast cells.
- Employed deep learning, specifically transfer learning with pre-trained convolutional neural network (CNN) models (VGG-19, ResNet50, Xception, DenseNet121) with adjusted classifiers.
- Compared the performance of these CNN models against other machine learning algorithms and visualized/interpreted the results.
Main Results:
- The study evaluated models on a dataset of 369 images containing 6112 cell cases.
- The DenseNet121 model achieved the highest performance metrics.
- DenseNet121 demonstrated 93.5% accuracy, 92% sensitivity, and 95% specificity in classifying cell morphology changes indicative of C. difficile toxin exposure.
Conclusions:
- Computer vision algorithms, particularly deep learning models like DenseNet121, can effectively identify morphological changes in cells exposed to C. difficile toxins.
- This automated approach offers a promising alternative to manual image analysis for accelerating drug discovery for CDI.
- The developed method shows high accuracy and specificity, suggesting its potential for identifying novel therapeutic candidates for CDI.
