Application of text mining to build AOP-based mucus hypersecretion genesets and validation with in vitro and clinical samples

Purpose: Build genesets using text mining and validate the geneset with data from public repository and a dataset generated in house using in vitro models and RNA-seq Methods: Mucus hypersecretion and mucociliary dysfucntion adverse outcome pathways (AOPs) genesets were build using the text mining method described by Rani et al., 2015. Validation was performed using RNA expression data from cigarette and e-cigarette aerosol treated cells, IL-13 treated airway cells, and COPD-lung biopsies. The cigarette and e-cigarette aerosol RNA-samples from airway cells were generated and sequenced in house. The other dataset were publically available. Results: Using unsupervised clustering, the mucus hypersecretion and mucociliary dysfunction genesets were able to discriminate the cigartte treated cells from the e-cigarettes and the air control. The e-cigarette and the air control clustered together. Clustering was also observed with IL-13 treated cells. IL-13 is an induced of mucus hypersecretion. Clustering was not observed when COPD RNA-seq samples were used. PCA analysis revealed some degree of grouping based on disease status, but this was also heavily confounded by other parameters. Conclusions: Our study described the first application of text mining to build genesets relevant to AOPs. In vitro validation confirmed the genesets could discriminates between treatment that induce mucus hypersecretion phenotypes, however this could not be confirmed with COPD biopsy samples. This could be due to a series of technical confouding factors and the heterogeneity of the COPD disease. Overall design: RNA was extracted from in vitro primary airway cells exposed at the air liquid interface to cigarette and e-cigarette aerosol. RNA-seq data was generated using an Illumina NextSeq 500 platform at a depth of 40 million pair reads with 150 bp paired-end