Prediction of longitudinal inflammatory phenotypes using baseline sputum transcriptomics in UBIOPRED

<b>Background:</b> 4 inflammatory phenotypes(paucigranulocytic P; eosinophilic E; neutrophilic N or mixed EN) were previously defined based on the sputum inflammatory cell counts in UBIOPRED severe asthma cohort[1]. These inflammatory phenotypes vary over time according to the effects of the environment or therapy. <b>Aim:</b> To determine whether sputum transcriptomics can predict the change in the inflammatory phenotype at 1 year. <b>Methods:</b> We defined EN as the most inflammatory group and the P as the least inflammatory group With E and N being intermediate. Sputum transcriptomics for (n=42) were labelled 1, 2, 3(see Table 1). A random forests classifier based on the Partial Least Squares components and leave-one-out cross-validation(cv) for testing with 3-fold cv for hyper-parameter tuning was used. EnrichR was used for pathway analysis. <b>Results:</b> The balanced accuracy of the classifier is 93 ±0.05% for Class 1 (n=23), 78±0.1% for Class 2 (n=8) and 78±0.1% for Class 3(n=11). Pathways are shown in Figure 1. <b>Conclusions:</b> Baseline sputum transcriptomics may be used to predict changes in inflammatory phenotype at 1 year. These findings need validation. Figure 1-Enriched biological processes for upregulated/downregulated genes pvalue &lt;=0.005, + stands for positive.