Illustration IA en science du vivant 2

Replay séance 3 : Silvia Bottini

Replay séance 3 : Silvia Bottini - Intégration de données multi-omiques grâce aux autoencodeurs variationnels

Cette 3ème séance, qui s'est déroulée le 6 février, a donné la parole à Silivia Bottini, récemment recrutée à l'Institut Sophia Antipolis, qui a présenté ses travaux sur l'Intégration de données multi-omiques grâce aux autoencodeurs variationnels.

Plants live in a constantly changing environment that is often unfavorable or even hostile. Indeed, they developed high phenotypic plasticity that includes rapid responses to aggressive biotic and abiotic factors and adaptations to changing environments. In case of multiple stresses occurring at the same time, plants activate appropriated signaling pathways to respond to both or by prioritizing the response to one stress over the other. While very few studies on plant response to multiple stresses at the same time have been reported, mainly because of the difficulties of the experimental design, several studies have been conducted to study one stress at the time. Currently used methods to integrate experiments conducted on single stress consist of performing meta-analysis or by finding differentially expressed genes for each experiment and then selecting the common ones.  Although these approaches allowed to find valuable results, they cannot identify non-specific conserved mechanisms that may hold promise for a broader understanding of plant defense response mechanisms.  Furthermore, very often, only few candidates are consistently found in agreement in the meta-analysis and usually the number decrease with the increasing of the number of experiments. 
For this purpose, we developed HIVE (Horizontal Integration analysis using Variational AutoEncoders (VAE)) to analyses horizontally integrated multi-omics datasets composed by unpaired and/or longitudinal experiments. Briefly, we coupled the VAE that captures non-linear relationships and encoded them in the latent space, with a random forest regression (RFR) and SHAP explainer to allow the explainability.
We illustrate the functionality of HIVE to study the transcriptional changes of two peanut wild relatives’ plants during root-knot nematode Meloidogyne arenaria infection and/or drought stress from six unpaired experiments. HIVE showed better performances compared to the meta-analysis and the state-of-the-art tool and identified novel promising candidates responsible for triggering effective defense response to biotic and abiotic stress.    

Voir le replay de la présentation de Silivia Bottini 

Date de modification : 04 mars 2024 | Date de création : 04 mars 2024 | Rédaction : Marjorie Domergue