Machine Learning Approaches to Assess Soil Microbiome Dynamics and Bio-Sustainability

Understanding soil microbiota dynamics is essential for enhancing bio-sustainability in agriculture, yet the complexity of microbial communities hampers the prediction of their functional roles. Artificial intelligence (AI) and machine learning (ML) offer powerful tools to analyse high-dimensional microbiome data generated by high-throughput sequencing. Here, we apply unsupervised AI-based algorithms to uncover microbial patterns that are not immediately recognisable but are crucial for characterising the biological status of agricultural soils. Soil samples were collected from a site in Northern Italy managed under four strategies: conventional farming without organic matter (C), with organic matter (C + O), with beneficial microorganisms but without organic matter (M), and with both beneficial microorganisms and organic matter (M + O). Metagenomic amplicon sequencing of the 16S ribosomal RNA (rRNA) gene and the internal transcribed spacer (ITS) region was used to profile bacterial and fungal communities. Principal component analysis (PCA), k-means clustering, and t-distributed stochastic neighbour embedding (t-SNE) revealed coherent temporal trajectories in both datasets, with sampling time and crop presence emerging as dominant drivers of community assembly and only subtle compositional shifts attributable to treatments. Fungal communities exhibited higher plasticity and a stronger response to management than bacterial communities, which converged towards a stable oligotrophic core. Our findings highlight the complementary roles of fungal and bacterial guilds and show that unsupervised ML-based workflows provide an effective framework to disentangle temporal and treatment effects in complex microbiome datasets. This exploratory study lays the groundwork for future predictive models aimed at identifying microbial indicators of soil biological status and supporting bio-sustainable agronomic decisions.