Zero inflated high dimensional compositional data with DeepInsight.

Area of Science:

• Microbiome Research
• Bioinformatics
• Computational Biology

Background:

• Human microbiome research has expanded due to advanced sequencing technologies like Next Generation Sequencing (NGS) and High-Throughput Sequencing (HTS).
• Microbiome data analysis frequently faces challenges with zero-inflation and high dimensionality, complicating compositional interpretation.
• Operational Taxonomic Units (OTUs) and Amplicon Sequence Variants (ASVs) are common numerical proxies used in microbiome studies.

Purpose of the Study:

• To develop a method for analyzing zero-inflated and high-dimensional microbiome data with compositional interpretation.
• To address the limitations of existing methods in accurately analyzing complex microbiome datasets.

Main Methods:

• Applied a square root transformation to compositional microbiome data to address zero-inflation, mapping it onto a hypersphere.
• Modified the DeepInsight image-generating method using Convolutional Neural Networks (CNNs) to accommodate hypersphere space for high-dimensional data.
• Introduced a strategy to differentiate true zeros from false zeros in zero-inflated data by adding a small value.

Main Results:

• The proposed method demonstrated effectiveness in analyzing pediatric inflammatory bowel disease (IBD) fecal sample data.
• Achieved an Area Under the Curve (AUC) of 0.847, surpassing the previous study's AUC of 0.83.

Conclusions:

• The developed approach successfully handles zero-inflated and high-dimensional microbiome data.
• This method offers improved diagnostic potential for diseases like IBD by enhancing microbiome data analysis.