Optimizing bioinformatic workflows to extract clinically usable gene expression data from targeted tumor RNA sequencing panels: comparison with total RNA-seq in cancer samples

Motivation Targeted RNA sequencing (RNA-seq) is widely used to detect gene fusions in tumors but clinical use of expression data from panels in fusion-negative cases has been limited. Differential gene expression profiling (DGE) from these panels has the potential to improve tumor classification. Results To facilitate this application, we compared methods for sequence read counting, gene normalization and supervised and unsupervised clustering methods to optimize them for smaller gene sets. We derived DGE data from ∼200-gene RNA-seq fusion panels. Among five tools for read counting, featureCounts was the most rapid and robust. For DGE with DESeq2, we compared five normalization strategies and showed the 5 most stably expressed genes over multiple sets provided optimal centralization. The outputs of the optimized pipeline were then assessed by a newly constructed targeted panel that added a limited number of genes assessing cell lineage and tumor grade. Finally, the optimized pipeline was evaluated using mean centroid and principal component analysis and pathway analysis and compared to outputs from full RNA-seq on a common set of challenging tumors. Comparable tumor clustering was observed with RNA-seq and the redesigned targeted gene panel. Availability and implementation The data analyzed during the current study are available from the corresponding author on reasonable request.