Deeptaxim: Comprehensive classification analysis for taxonomic datasets using image-based deep-learning models.

Area of Science:

• Microbiology
• Bioinformatics
• Artificial Intelligence

Background:

• Microbiome data presents challenges in classification due to high dimensionality and variability.
• Deep learning offers potential solutions for analyzing complex biological datasets.

Purpose of the Study:

• To apply deep learning for accurate microbiome data classification.
• To develop and assess the Deeptaxim model for disease prediction using microbiome data.
• To evaluate transfer learning capabilities and robustness of the Deeptaxim model.

Main Methods:

• Microbiome data converted into image format using taxonomic cladograms.
• 2D-CNN-based Autoencoder, U-Net, and GAN architectures utilized in the Deeptaxim model.
• Comparison of image-based classification with conventional methods.

Main Results:

• Taxa-ordered images and CNN classifiers outperformed traditional methods for microbiome data classification.
• Transfer learning significantly improved classification performance on low-sample and varied disease datasets.
• The Deeptaxim model demonstrated robustness across different disease types.

Conclusions:

• Image-based representation of microbiome data enhances deep learning classification accuracy.
• Deeptaxim shows promise for disease prediction and can be adapted as a wellness index.
• The model's NN-based framework allows integration into other deep learning architectures.