Engineering synthetic microbial communities for soil restoration: from rational design to multi-scale biogeochemical applications

Abstract

Soil degradation caused by intensive land use, pollution, and climate stress threatens food production and ecosystem function. Single-strain inoculants often perform inconsistently because they do not persist or compete well in resident soil microbiomes. Synthetic microbial communities (SynComs) combine complementary microbial functions in defined consortia. Their performance, however, depends on community composition and environmental context. Multi-omics, high-throughput screening, and artificial intelligence (AI) can reduce the number of strain combinations that need to be tested. Yet the roles of AI and other computational approaches are often described imprecisely. We distinguish four roles: direct AI design, AI-assisted candidate discovery, model-guided design, and prediction-only analysis. This framework links computational methods to the experimental steps needed to construct and validate soil SynComs. We examine applications in nutrient acquisition, carbon cycling, pollutant remediation, disease suppression, and tolerance to drought and salinity. Direct AI design has so far been demonstrated only in a few controlled plant systems and soil microcosms. In most studies, machine learning (ML) is used earlier in the workflow to identify candidates, while cultivation, functional assays, and interaction tests determine the final community composition. To bridge this gap, we propose a staged design-build-test-learn (DBTL) framework, which links model evaluation and community reconstruction with testing in non-sterile soil, strain tracking, safety assessment, and field validation. AI can narrow the experimental search space, but empirical validation remains essential to establish causality, persistence, and transferability across soils and hosts.