Automatic classification of fungal-fungal interactions using deep leaning models

Area of Science:

• Agricultural Science
• Biotechnology
• Computer Science

Background:

• Fungi are crucial for biotechnology, offering solutions in food, healthcare, and agriculture.
• Current methods for identifying biocontrol fungi are subjective, time-consuming, and lack reproducibility.

Purpose of the Study:

• To develop an automated image classification approach using deep learning for analyzing fungal-fungal interactions.
• To overcome the limitations of traditional subjective observation methods in biocontrol fungi identification.

Main Methods:

• Utilized a deep neural network-based machine learning algorithm for image classification.
• Analyzed standardized images of 96-well microtiter plates from fungal-fungal challenge experiments.
• Trained and evaluated multiple deep learning architectures on interactions involving *Fusarium graminearum*.

Main Results:

• Achieved a peak accuracy of 95.0% using the DenseNet121 model.
• Obtained a maximum macro-averaged F1-Score of 93.1 across five folds.
• Demonstrated the effectiveness of the AI approach in classifying fungal interactions.

Conclusions:

• The developed AI-automated method provides an efficient and reproducible way to classify fungal-fungal interactions.
• This deep learning approach can be adapted for identifying other fungal species and applications.
• The study presents the first automated deep learning method for classifying fungal-fungal interactions.