Ordering taxa in image convolution networks improves microbiome-based machine learning accuracy.

Area of Science:

• Microbiology
• Bioinformatics
• Machine Learning

Background:

• The human gut microbiome is linked to numerous diseases, making it a target for machine learning (ML)-based biomarker development.
• Microbial sequence-based studies face challenges for ML, including data sparsity, high dimensionality, and non-uniform representation.
• Current methods struggle with the inherent complexities of microbiome data, limiting accurate ML applications.

Purpose of the Study:

• To develop a novel method for improving machine learning applications in microbiome research.
• To enhance the representation and analysis of microbial taxonomy for more accurate disease biomarker identification.
• To create an interpretable ML framework for understanding microbiome-disease associations.

Main Methods:

• A graph representation was used to show cladogram structure is as informative as taxa frequency.
• iMic (image microbiome) translates microbiome data into images using an iterative ordering scheme.
• Convolutional neural networks (CNNs) were applied to the generated images, and explainable AI was used for interpretation.

Main Results:

• iMic demonstrated higher precision in static microbiome gene sequence-based ML compared to state-of-the-art methods.
• The method effectively combines information from different taxa, improving data representation for ML.
• Explainable AI facilitated the identification of taxa relevant to specific disease conditions.

Conclusions:

• iMic offers a powerful and interpretable approach for microbiome data analysis and biomarker discovery.
• Translating microbiome data into images significantly enhances ML model performance.
• The iMic framework can be extended to analyze dynamic microbiome samples, opening new avenues for research.