A merged lung cancer transcriptome dataset for clinical predictive modeling

Abstract

The Gene Expression Omnibus (GEO) database is an excellent public source of whole transcriptomic profiles of multiple cancers. The main challenge is the limited accessibility of such large-scale genomic data to people without a background in bioinformatics or computer science. This presents difficulties in data analysis, sharing and visualization. Here, we present an integrated bioinformatics pipeline and a normalized dataset that has been preprocessed using a robust statistical methodology; allowing others to perform large-scale meta-analysis, without having to conduct time-consuming data mining and statistical correction. Comprising 1,118 patient-derived samples, the normalized dataset includes primary non-small cell lung cancer (NSCLC) tumors and paired normal lung tissues from ten independent GEO datasets, facilitating differential expression analysis. The data has been merged, normalized, batch effect-corrected and filtered for genes with low variance via multiple open source R packages integrated into our workflow. Overall this dataset (with associated clinical metadata) better represents the diseased population and serves as a powerful tool for early predictive biomarker discovery.