Modelling single-stress omics integration with HIVE enables the identification of responding signatures to multifactorial stress combinations in plants

All organisms are subjected to multiple stresses usually occurring at the same time, requiring the activation of the appropriate signalling pathways to respond to all or by prioritizing the response to one stress factor. Plants, as sessile organisms, are particularly impacted by the constantly changing environment that is often unfavourable or even hostile. Because of the experimental complexity of studying the response of one organism to multiple stressors simultaneously, usually experiments are conducted considering one individual stress factor at the time. An alternative consists in performing in silico integration of those data on single stress response. Currently used methods to integrate unpaired experiments consist of performing meta-analysis or finding differentially expressed genes for each condition separately and then selecting the commonly regulated ones. Although these approaches allowed to find valuable results, they mainly identify specific signatures in response to one stress and very few signature responding to multiple stresses and lack those modulated differently in each condition. For this purpose, we developed HIVE (Horizontal Integration analysis using Variational AutoEncoders) to integrate multiple single-stress transcriptomics datasets composed of unpaired experiments. Briefly, we coupled a variational autoencoder, that alleviates batch effects, with a random forest regression and the SHAP explainer to select relevant genes modulated specifically in response to one or multiple stresses. We illustrate the functionality of HIVE to study the transcriptional changes of several different plants namely Arabidopsis thaliana, rice, maize, wheat, grapevine and peanut by collecting publicly available experiments on single stress, either biotic and/or abiotic, and jointly analyse them. HIVE performed better than the differential expression analysis, meta-analysis and the state-of-the-art tool for horizontal integration allowing to identify novel promising candidates responsible for triggering effective defence responses to multiple stresses.