MYCOLab Ai Logo - gold and army green

Automatic classification of fungal-fungal interactions using deep learning models

Summary

Researchers developed an artificial intelligence system that automatically analyzes images of fungi growing together to identify which ones can fight off disease-causing fungi. Instead of having humans manually look at thousands of plate images, which is time-consuming and subjective, their computer vision system can classify the outcomes with 95% accuracy. This automation tool could help scientists quickly find beneficial fungi that could replace chemical pesticides in agriculture, supporting the goal of sustainable and more environmentally friendly farming.

Background

Fungi are valuable sources for biotechnological applications including enzymes, bioactive metabolites, and biocontrol organisms. Current workflows for identifying new biocontrol fungi rely on subjective visual observations, making the process time-consuming and difficult to reproduce. This study addresses the need for automated classification of fungal-fungal interactions using artificial intelligence.

Objective

To develop an automated image classification approach using deep neural networks to analyze standardized images of 96-well microtiter plates for fungal-fungal challenge experiments. The study focused on categorizing interactions between the plant pathogen Fusarium graminearum and fungal isolates from a collection of 38,400 strains.

Results

DenseNet121 achieved the highest performance with 95.0% accuracy, 92.4% precision, 94.1% recall, and 93.1% F1-score. ResNet50 and ViT also demonstrated strong performance with 94.9% and 94.5% accuracy respectively. All models showed robust performance across five-fold cross-validation, effectively classifying fungal interactions as candidate wins, pathogen wins, or draws.

Conclusion

This study introduces the first automated method for classifying fungal-fungal interactions using deep learning, successfully addressing the limitations of manual inspection. The approach reduces subjectivity, increases reproducibility, and can be adapted for other fungal species, providing a valuable tool for identifying biocontrol organisms and supporting sustainable agriculture.
  • Published in:Computational and Structural Biotechnology Journal,
  • Study Type:Experimental Study,
  • Source: 10.1016/j.csbj.2024.11.027, PMID: 39655263
Scroll to Top