A t-SNE Based Classification Approach to Compositional Microbiome Data.

Area of Science:

• Microbiome analysis
• Bioinformatics
• Machine learning

Background:

• t-distributed stochastic neighborhood embedding (t-SNE) is a powerful tool for dimensionality reduction and visualization.
• t-SNE is increasingly utilized for classification and regression tasks.
• Microbiome data is compositional, requiring specialized analytical approaches.

Purpose of the Study:

• To develop a t-SNE based classification approach for compositional microbiome data.
• To adapt t-SNE to account for the unique characteristics of microbiome data.
• To evaluate the performance of the proposed t-SNE classification method.

Main Methods:

• Modified t-SNE by incorporating Aitchison distance to handle compositional data.
• Classified new samples using low-dimensional features derived from nearest neighbors in the reduced space.
• Employed logistic regression (LR), support vector machine (SVM), and decision tree (DT) for classification.

Main Results:

• The t-SNE based classification approach achieved superior performance compared to models built in the original high-dimensional space.
• Incorporating Aitchison distance into t-SNE improved classification accuracy on two disease-associated microbiome datasets.
• The developed method effectively classifies new samples within the t-SNE reduced dimensional space.

Conclusions:

• A novel t-SNE based classification approach suitable for compositional microbiome data has been developed.
• The proposed method demonstrates improved classification accuracy and offers a valuable baseline for future microbiome classification models.
• This approach enhances the utility of t-SNE in analyzing complex biological datasets.