MicroSTNet: a spatio-temporal graph-based framework for time-series microbiome analysis.

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Microbial community structure and function are influenced by spatio-temporal dynamics.
• Current machine learning models for phenotype prediction often overlook these dynamics, limiting accuracy, especially with single time-point data.

Purpose of the Study:

• To investigate microbial community interaction dynamics in closed environments.
• To develop and validate a novel model for predicting dynamic microbial abundance and future community trends.

Main Methods:

• Introduced a microbial spatio-temporal network model combining two-stream spatio-temporal graph convolutional networks and long short-term memory.
• Applied the model to predict microbial abundance in the human oral cavity and gut using data from two independent projects.

Main Results:

• The model accurately captures temporal trajectories and spatial network features of microbial communities.
• Experimental validation confirmed high accuracy in tracking temporal patterns, even for fluctuating microorganisms.
• Ablation studies showed the integrated model outperforms individual components.

Conclusions:

• The microbial spatio-temporal network model effectively predicts dynamic microbial abundance and future community trends.
• This technology offers a promising approach for low-cost, non-invasive early disease diagnosis and health risk assessment.