NetGAM: Using generalized additive models to improve the predictive power of ecological network analyses constructed...

Area of Science:

• Ecology
• Microbial ecology
• Bioinformatics

Background:

• Ecological network analysis infers microbial interactions from species abundance data.
• Time-series data present statistical challenges, potentially leading to inaccurate network predictions due to non-biotic associations.

Purpose of the Study:

• To develop and validate a data transformation method to remove time-series signals from microbial abundance data.
• To improve the accuracy and reliability of ecological network inference.

Main Methods:

• A generalized additive model (GAM)-based transformation was applied to remove temporal signals from species abundance data.
• Mock time-series datasets with known covariance structures were used for validation.
• Network analyses were performed with and without the GAM transformation, comparing outputs to the known structure.

Main Results:

• Seasonal abundance patterns significantly reduced the accuracy of inferred ecological networks.
• The GAM transformation enhanced the predictive power (F1 score) of ecological networks.
• The transformation improved the ability of network inference methods to capture key network structures.

Conclusions:

• Temporal dynamics, such as seasonal patterns, must be addressed in ecological network analyses.
• The described GAM transformation is a simple and effective tool for improving the accuracy of ecological network inference from time-series data.