Current ongoing projects :
Axis 1.
HIVE: a novel method to analyse unpaired multi-transcriptomics data
Plants live in a constantly changing environment in which multiple stresses can occur at the same time, requiring the plants to activate appropriate signaling pathways to respond to these stresses. However, it is very complex to reproduce experimentally multi-stress conditions. An alternative is to integrate in silico omics experiments performed in systems each studying one single stress. Multiple challenges need to be addressed to be able to jointly analyse data from different experiments. Batch effects are common technical variations in omics and multi-omics data and lead to equivocal results if uncorrected or over-corrected. To tackle those challenges, we have recently developed HIVE, a method based on variational autoencoders, random forest regression and SHAP explainer to jointly analyse multiple transcriptomics data from different experiments or batches.
PIVAE: Integrating Physics-Informed Neural Networks and Variational Autoencoders for Enhanced Omics Data Interpretation in Biological Applications
Plant defense against biotic threats requires multiple signaling processes that are influenced by varying spatial and temporal factors, at different biological layers, altogether concurring for a successful or unsuccessful infection. Furthermore, timely and rapid plant response to these attacks is essential and can dramatically affects plants fate. Therefore, tracking when particular immune responses are induced accordingly to different pathogen infection is crucial for understanding plant immunity. Despite often longitudinal experiments are performed, current available analysis methods do not explicitly consider the time dependency of successive observations. We aim to leverage Physics-Informed Neural Networks (PINNs) and Variational Autoencoders (VAEs) for parameter estimation in systems where the underlying physical dynamics are unknown or partially understood. The ultimate goal is to enhance the interpretation and integration of omics data within biological systems.
Development of a novel machine learning approach for the integrated analysis of transcriptome, degradome and miRNAome
Among the regulation mechanisms, microRNAs (miRNAs) play an important role in eukaryotes at both transcriptional and post-transcriptonal levels. These regulatory processes are complex to decipher because one miRNA can have multiple targets and one target can be targeted by multiple miRNAs. To get insight into these mechanisms, degradomics can be used which aims at quantifying target degradation by identifying miRNA cleavage sites. However, current methods used to analyze these data are only limited to find the specific pattern of reverse regulation to highlight an up-regulated differentially expressed miRNA that targets a down-regulated differentially expressed transcript. To attempt to have a more global understanding of the mechanisms regulating miRNA-target relationships, we will develop a new deep learning model to analyze transcriptome, microRNAome and degradome in an integrative perspective.
Axis 2.
TomTom: a knowledge graph for tomato molecular interactions
Leveraging the collaboration with Dr. Aurelien Dugourd at the University of Heidelberg we are constructing a knowledge graph for tomato interactions. TomTom represents a fingerprint of a wide type of molecule interactions comprising 113 415 nodes and 2 864 036 relationships represented as a neo4j graph database using Biocypher. The interrogation of TomTom enables interactive exploration of current knowledge and will help to provide mechanistic hypotheses and new insight into experimental observation.
Novel models based on the topological data analysis to identify robust molecular signatures defining plant cell fate upon pest aggression.
In collaboration with Dr. Carriere at Inria Sophia-Antipolis, we apply and develop novel models in the framework of the topological data analysis to improve the analysis of gene regulatory networks in the context of plant-pathogens interactions.
Axis 3.
Disentangling plant response to biotic and abiotic stress using Arachis species as a model
In collaboration with Dr. Ana-Paula Zotta-Mota from team GAME at ISA and Dr. Patricia Guimares at EMBRAPA in Brazil we studied the transcriptional changes of two Arachis wild species submitted to root-knot nematode Meloidogyne arenaria infection and/or drought stress from seven unpaired experiments. The application of HIVE enabled to select common and specific molecular signatures encompassing the different plant species and stresses. Furthermore, to better understand the crosstalk in response to biotic and abiotic stress conditions, we reconstruct the gene regulatory network. The study of the inferred network allowed to identify two NBS-LRR with a potential role to confer resistance against RKN infection.
Identifying specific and/or shared multi-omics signatures of the response of Arabidopsis thaliana to multiple biotic agents
Exploiting the collaboration with Drs. Harald Keller, Agnes Attard and Dr. Bruno Favery from IPO and IPN team at ISA, respectively, we focused on the response of Arabidopsis thaliana to multiple biotic agents, to characterise the molecular mechanisms involved in a broad spectrum of plant-pathogen interactions. The project focused on interactions with foliar and root pathogens, and the flagellin 22 peptides, a component of PAMPs. For foliar pathogens, the oomycete Hyaloperonospora arabidopsidis virulent strain, Noco2 and an avirulent strain, Emwa1 and the fungus Alternaria brassicicola, were selected. For root pathogens, we chose to characterise the response to the nematode Meloidogyne incognita and the oomycete Phytophthora parasitica. To understand this complexity of responses of A. thaliana to those pathogens, we use a multi-omics approach, with transcriptomic (study of transcripts) and translatomic (study of transcripts in the process of being translated) data, coupled with co-expression network strategy.
POMOdORO: Pan OMics cOllection of tOmato undeR biOtic stress
In collaboration with Eng. Corinne Rancurel and Martine Da-Rocha at the bioinformatic platform at ISA we conducted an unprecedent literature mining study to collect omics data evaluating multiple biotic perturbation of S. lycopersicum. Several online databases were interrogated to retrieve several publications to include in our database, for a total of 12 years of publication, five type of omics data, from 17 countries all around the world. This collection of more than 2000 samples across 39 distinct pathogens interactions, several developmental stages, sampled tissues and tomato varieties will allow to investigate how tomato plant reacts to a broad spectrum of biotic agents from ectoparasites to endoparasites, from biotrophic to necrotrophic, from avirulent to virulent strains, attacking different tissues from leaves to roots, causing different possible diseases, yielding a pan-tomato-biotic repertoire of functionally relevant impacted molecules.
Understanding the plant-pathogen molecular dialogue during the infection by studying the gene regulatory network in tomato
To study the response of tomato to multiple pathogens, we have selected a subset of six omics datasets from five studies from the data collected in POMOdORO, and applied HIVE and TomTom to reconstruct and study the gene regulatory network. By crossing topological considerations and estimation of transcription factor activity, we identified novel and well-known key regulators which are activated either in response to specific infection or upon several attacks but with different regulatory consequences. This study is a proof of concept to study plant multi-pathogens interactions.
