Difficulty in inferring microbial community structure based on co-occurrence network approaches.

Area of Science:

• Microbial Ecology
• Bioinformatics
• Computational Biology

Background:

• Co-occurrence networks infer ecological associations from microbial community sequencing data.
• These networks are widely used to study microbial interactions but their validity is debated.
• Current methods often rely on parametric models, which may not suit microbial population dynamics.

Purpose of the Study:

• To comprehensively evaluate the validity of common co-occurrence network inference methods.
• To assess how well these methods describe interaction patterns in simulated ecological communities.
• To compare the performance of co-occurrence network methods against classical statistical approaches.

Main Methods:

• Realistic simulations of ecological communities were employed.
• Nine widely used co-occurrence network inference methods were evaluated.
• Performance was assessed based on the accuracy of interaction pattern reconstruction.

Main Results:

• Co-occurrence network methods performed similarly to or worse than classical methods like Pearson's correlation on compositional data.
• The methods inadequately described interaction patterns in dense and/or heterogeneous networks.
• Performance varied by interaction type, with competitive interactions predicted more accurately than predator-prey interactions.

Conclusions:

• Co-occurrence network approaches may be insufficient for interpreting species interactions in microbiome studies.
• Further rigorous evaluation of existing methods is necessary.
• Development of more suitable methods for inferring microbial ecological networks is needed.