Da floresta secundária à agricultura: machine learning e bioinformática na avaliação dos efeitos do uso do solo sobre o microbioma edáfico no estado do Paraná

Abstract

The soil is home to a complex microbial community that is essential for maintaining ecosystem services and agricultural sustainability. The conversion of natural ecosystems into agricultural areas drastically alters edaphic properties and soil microbiota. To understand these changes, it is essential to integrate traditional ecology with high-resolution analytical approaches. This study aimed to propose an analytical workflow, combining multivariate statistics and machine learning, to model bacterial structuring and infer its edaphic drivers. Under the hypothesis that predictive models reveal complex ecological signatures of anthropogenic management, edaphic data integrated with 16S rRNA sequencing from four locations in the state of Paraná were used. Bioinformatics tools (DADA2), Variation Partitioning, and computational algorithms were employed to evaluate the influence of land use on the microbiota. The models captured specific ecological signatures for each management type, demonstrating that biological attributes are the main drivers of microbial composition. A strong coupling was evidenced between agricultural systems and copiotrophic phyla (Pseudomonadota), contrasting with the retention of sensitive oligotrophs (Acidobacteriota and Verrucomicrobiota) in the Secondary Forest (SF). Topological network analysis indicated that the SF exhibits greater complexity and strong biochemical coupling, sustaining the highest diversity of detectable ASVs. Considering the computational detection limit in agricultural systems, the forest acts as a reservoir and a dynamic ecotone. However, its functional effectiveness in the landscape depends on spatial scaling, since diminutive fragments suffer from edge effects and biological homogenization. It is concluded that integrating machine learning and multivariate statistics is a robust approach for identifying patterns of ecological stability. It is inferred that the resilience of the evaluated agroecosystems is closely linked to the maintenance of conservation practices and adequate forest remnants. This work provides an essential analytical workflow for the future development of edaphic bioindicators.