Abstract
Ensuring food security for a population that will exceed 9 billion people by 2050 while preserving the environment and biodiversity in the context of environmental degradation and climate change, is one of the major challenges of our society. All species, from microorganisms to plants, are engaged in various biotic and abiotic interactions that can benefit their growth or harm their health and survival. In extreme conditions, abiotic factors like temperature, water availability, wind, and UV radiation can become deleterious for the ecosystem. Accordingly, the uncontrolled proliferation of non-beneficial biotic factors like pathogens will cause severe outbreaks. Adaptation, especially for sessile organisms, is therefore fundamental. Indeed, advancing our knowledge on the pathogenic biotic and extreme abiotic factors interacting with living organisms is mainstay. In agriculture, plant and phytopathogen interaction is of major interest in developing plant protection measures. Typically, plant parasitic organisms like fungi, oomycetes, nematodes, and phytoplasmas encode specific pathogenicity factors called effector proteins responsible for the disease's development. Advances in biotechnology especially with the development of the so‐called omics techniques and their spectacular progress during the last two decades, have led to the possibility of gaining insightful knowledge with which answer to complex biological questions. Omics approaches aim to understand a complex biological system as a whole, applying systematic analysis of its content at the molecular level. On the other hand, several challenges are entangled with omics data, as high data dimensionality and heterogeneity, requiring the use of high-throughput bioinformatics and artificial intelligence models to extract biological meaning. Despite the incredible amount of tools and models available for omics data analysis with very successful results in several field of biology, there is still the need for tailored methods that account for the complex interaction of biotic and abiotic stresses with plants and microbial organisms. In this thesis, I studied a wide variety of organisms from fungi, plant-pathogens such as oomycetes, nematodes and phytoplasma bacteria, to apples and two wild peanuts plant species as biological model to investigate the pivotal characteristics of biotic and abiotic factor responses. First, I focused on the study of the adaptation of a peculiar group of fungi called black fungi or Rock Inhabiting Fungi by comparing two major classes, Dothideomycetes and Eurotiomycetes. In this context, I applied statistical methods and machine learning to find each class's most related metabolic competence and the significant differences in response to the preferred environment (Chapter I). After that, as faces of the same medal, I explored the phytopathogenic organisms as biotic stresses and the relative plant defense response, to investigate the complex dialog of these organisms in the race for survival, considering their role in important agroeconomic losses. I first characterized the physicochemical features of effector proteins, as major pathogenicity factors in causing the disease, by developing a bioinformatic tool, MOnSTER, to find clusters of protein motifs strongly related to this protein class and ranked by significative relationship. The tool was designed to find clusters of already known effector motifs in plant pathogenic oomycetes and then applied to newly discovered effector protein motifs in Plant Parasitic Nematodes (Chapter II). I then used these results and other essential features to implement an ensembled machine learning model for the prediction of effector proteins in the ‘Candidatus Phytoplasma’ genus, called LEAPH. Along with the predictor comes a self-organizing map that depicts the putatively predicted and experimentally validated effector landscape of 13 phytoplasma proteomes. This will contribute to increase the knowledge about these still poorly characterized proteins and their specific role in the symptom’s development. The latest aim is therefore to boost the assessment of possible countermeasures in favor of plant health and environmental preservation (Chapter III). To investigate the plant defense response, I first focused on a controlled experimental design in which each plant lineage, in this case, the apple plant, is maintained in vitro and infected by a single strain of 'Candidatus Phytoplasma mali' (Chapter IV). Given the obligate parasitic nature of phytoplasmas that cannot survive outside of their hosts, I initially implemented a novel automated assembly pipeline, PLANTOGEN-A, able to assemble good quality pathogen genome from the sequencing data of both the host and the parasite. Then I compared the expression profiles of differently infected plants to understand how pathogens modulate the plant response both genetically and phenotypically. Afterward, to explore a more physiological condition in which plants are subjected to different biotic and abiotic stresses simultaneously, I implemented a tool, HIVE, that leverages a deep learning algorithm, a variational autoencoder, to integrate unpaired multi-batch transcriptomic studies. I applied HIVE to study the transcriptome profiles of two wild peanut plants subjected to biotic or abiotic stresses to automatically detect which genes are majorly involved in the response to one stress or the other or the two simultaneously (Chapter V). In summary, I developed several computational tools at the crossroad between bioinformatics and machine learning tailored for agronomy and environmental sciences. Altogether, these studies have provided a whole perception of the complex interaction between biotic, abiotic, microbial communities and plant interaction that will help elaborate more effective and safe approaches to plant protection and environment preservation.