Reorganizing heterogeneous information from host-microbe interaction reveals innate associations among samples

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Operational taxonomic unit (OTU) abundance tables contain valuable host-microbe interaction data.
• Co-occurrence networks have limitations in capturing the full complexity of OTU data due to sparsity.
• Existing methods inadequately leverage sample-OTU interaction information.

Purpose of the Study:

• To develop a method for extracting more information from OTU abundance tables.
• To represent biological characteristics of OTUs and samples quantitatively.
• To improve the prediction of sample attributes using microbiome data.

Main Methods:

• Constructed a sample-OTU heterogeneous information network.
• Applied heterogeneous graph embedding to create OTU and sample spaces.
• Developed the Integrated Model of Embedded Taxonomies and Abundance (IMETA) using embedded vectors and abundance data.

Main Results:

• The generated OTU and sample spaces contain meaningful biological and medical semantic information.
• The IMETA model demonstrated stable and effective performance in sample classification tasks.
• The embedding representation based on the sample-OTU network provides richer insights into microbiome samples.

Conclusions:

• The proposed method significantly increases the utilization of information within OTU abundance tables.
• This approach offers a deeper understanding of the human microbiome.
• Quantified representations of biological characteristics enhance microbiome data analysis.