Data, machine learning models weights, and results used for the analyses in the paper entitled "Predicting microbial community structure and temporal dynamics by using graph neural network models".

Abstract from the manuscript:<br>Modeling or even predicting the dynamics of complex microbial communities comprising thousands of individual species has been a long-standing challenge within microbial ecology. In biological wastewater treatment plants (WWTPs) an optimal microbial community composition is critical to effectively remove or recycle resources from human activities that would otherwise pollute the environment. Both the presence and abundance of process-critical microorganisms are important for the overall performance, and the relative abundance of the individual species can vary greatly over time without any recurring abundance dynamics. Being able to predict the future abundance dynamics can therefore be a great help for WWTP operators to prevent problems in time and optimize operation and performance. Here we used a graph neural network based model design that, for the first time, enables accurate prediction of the future abundance dynamics of the majority of bacteria in the microbial communities of activated sludge several months ahead. We trained and tested models on time series datasets from 24 different Danish full-scale WWTPs sampled 2-5 times a month over a period of 3-8 years totalling 4709 environmental samples. We were able to predict the dynamics 10 time points into the future (2-4 months) with high accuracy for most of the abundant bacteria, especially process-critical bacteria, and some even 20 time points ahead (up to 8 months). The approach was implemented as a workflow suitable for any longitudinal community dataset, and is not limited to the AS ecosystem alone.<br>