Hybrid Soil Microbiome Modeling - Combining process-based models with machine learning to predict microbial dynamics and organic matter turnover in soil systems

Soil microorganisms control organic matter cycling and largely determine how soil systems can cope with and mitigate climate change and environmental threats. Integrating microbial dynamics in process-based soil models is critical for predicting how soil carbon flows and stocks change in ecosystems with time. Functional traits can be inferred from amplicon sequencing data and metagenome assembled genomes to leverage model parameterization. However, informing models using omics-based datasets is challenging due to their large dimensional nature and the nonlinear relationship between genomes and the actual function microbes express. We present a hybrid modeling framework that combines machine learning to analyze metagenomic and DNA sequencing data with a simple microbial explicit process-based model. This hybrid model is conditioned using a convolutional network trained with data from the LUCAS 2018 database (Land Use and Coverage Area frame Survey), which includes soil metagenomes, 16S sequencing data in combination with soil carbon, microbial biomass and soil respiration measurements. Using trait inference from genomes, the model can learn several biokinetic parameters such as growth rates, dormancy rates, affinities to organic matter, growth yields or decay rates. We present the concept of the hybrid soil modelling framework and discuss what data is informative for these models and how to best link machine learning with process-based models.