Microbial Community Prediction in Plant-Soil Systems Using Machine Learning

The integration of machine learning (ML) into microbiome research has opened new forontiers for analyzing microbial community data, and only recently has it been applied to better understand soil-plant interactions. Statistical analysis of amplicon and shotgun metagenomic sequencing data has long favored tree-based algorithms due to their ability to handle compositional, sparse, and high-dimensional data, even when sample sizes are limited. However, the application of deep learning, though often reserved for large datasets, can also be advantageous in soil and plant microbial ecology if the risk of overfitting is properly managed.

In this study, we used ML tools to investigate the extent to which different soil environments influence the composition of endophytic bacterial communities in Setaria viridis, a model species for C4 grasses. Seeds from two distinct locations in Austria were grown in both native and non-native soils, and microbial communities from various plant compartments were analyzed using 16S rRNA sequencing. The results show that soil plays a significant role in shaping these communities, with distinct microbial shifts observed when plants were grown in non-native soil. Nevertheless, a core set of seedborne bacteria persisted.

By using ML algorithms to analyze Amplicon Sequence Variants (ASVs), we generated predictive models supporting the hypothesis that soil exerts a driving influence on plant-associated microbiota while retaining some transmitted endophytes. By employing a variety of algorithms—from tree-based models to neural networks—we determined that certain soils drive more significant shifts in the microbial composition than others, offering valuable insights into the soil’s role in shaping plant microbiomes under varying environmental conditions.