Pace, Roberta (2025) Artificial Intelligence approaches to explore Soil Microbiome Dynamics: predicting soil temperature from soil microbial community profiles with a Feed-Forward Artificial Neural Network (ANN). [Tesi di dottorato]

Documento PDF Pace_Roberta_38.pdf Download (14MB)

Abstract

Soil represents a fundamental component of terrestrial ecosystems and a cornerstone of agricultural productivity. Its health and fertility are largely governed by the complex interactions among microbial communities that drive nutrient cycling, organic matter turnover, and ecosystem resilience. However, understanding these interactions and their responses to agronomic management remains a major scientific challenge due to the high complexity and multidimensionality of microbiome data. This PhD research, conducted within the National PhD in Artificial Intelligence (Agrifood and Environment area) at the Institute for Sustainable Plant Protection (CNR-IPSP, Portici, Italy), aimed to investigate soil microbial dynamics under different agronomic treatments by integrating metagenomic sequencing and AI-based analytical frameworks. Experimental work was carried out in a greenhouse cultivation of Valerianella locusta in Bagnolo Mella (Brescia, Northern Italy), where four treatments were applied: (i) conventional farming without organic matter (C), (ii) conventional farming with organic matter (C+O), (iii) conventional farming with beneficial microorganisms (M), and (iv) conventional farming with both microorganisms and organic matter (M+O). Soil samples collected at multiple time points were analysed using Oxford Nanopore sequencing (16S rRNA and ITS markers). Data processing and analysis were performed through Python-based pipelines, combining ML approaches to uncover patterns of microbial succession. Results revealed clear temporal-driven trajectories in both bacterial and fungal communities, and in some cases, with organic matter and microbial inoculation acting synergistically to enhance diversity and functional redundancy. Fungal communities showed higher sensitivity to treatments and temporal changes, while bacterial communities displayed greater stability and ecological resilience. The research outcomes include a review article published in SN Applied Sciences and an original research article submitted to Physiologia Plantarum for the Special Issue “Smart Agriculture – BrIAS edition” (Brussels Institute for Advanced Studies), linked to the II Conference on Smart Agriculture (Brussels, 2025). Additional analyses developed within this thesis further integrated the results through ANN, reinforcing the predictive potential of microbial community data. Ultimately, this work demonstrates how AI-driven approaches can bridge metagenomics and soil ecology, providing a methodological foundation for the development of predictive models of soil health. The results contribute to the broader goal of designing data-informed, sustainable agricultural systems where soil biodiversity is not only monitored but actively leveraged to ensure productivity, resilience, and environmental integrity.

Downloads

Actions (login required)

Modifica documento