Cross-validation for training and testing co-occurrence network inference algorithms.

Area of Science:

• Microbial ecology
• Bioinformatics
• Computational biology

Background:

• Microorganisms inhabit diverse environments and play vital roles in ecological processes and host health.
• Co-occurrence network inference algorithms are essential for understanding complex microbial associations, particularly in bacteria.
• High-throughput sequencing generates vast microbiome data, demanding robust computational methods for network analysis.

Purpose of the Study:

• To develop and validate a novel cross-validation method for evaluating co-occurrence network inference algorithms.
• To introduce new methods for applying existing algorithms to predict on test data.
• To address limitations of existing network evaluation techniques in microbiome research.

Main Methods:

• Proposed a novel cross-validation framework for network inference evaluation.
• Developed methods for applying existing algorithms to predict on test data.
• Focused on handling compositional data and high-dimensional, sparse microbiome datasets.

Main Results:

• The novel method demonstrates superior performance in handling compositional data.
• The framework effectively addresses challenges of high dimensionality and sparsity in microbiome datasets.
• Robust estimates of network stability were achieved.

Conclusions:

• The proposed cross-validation method is effective for hyper-parameter selection and comparing network inference algorithms.
• This advancement provides a reliable tool for analyzing complex microbial interactions.
• The framework establishes a new standard for network inference validation, applicable to various fields beyond microbiome studies.