Computational tool incorporates transcriptome size to improve RNA-seq analysis

Researchers studying gene expression have access to vast amounts of data from cells or tissues. This is thanks to advances in bulk and single-cell RNA-sequencing (RNA-seq) technologies that can capture every RNA molecule expressed in a cell: its transcriptome.

The computational methods developed to sift through these large transcriptomic datasets have led to groundbreaking discoveries, but they may also overlook data in ways that could mask meaningful findings. To get more out of RNA-seq analysis, scientists at St. Jude Children’s Research Hospital created ReDeconv, an algorithmic tool that accounts for transcriptome size differences to unmask those meaningful results.

Detailed information about the ReDeconv tool was published Feb. 1 in Nature Communications.

Genes are expressed when messenger RNA is made from DNA, which is then translated into proteins. Researchers use single-cell and bulk RNA-seq analysis to study gene expression, creating vast amounts of data. To deal with those large datasets, scientists use techniques that incorporate mathematical conventions to streamline and accelerate processing.

While effective, these conventions often prioritize computational efficiency over biological accuracy, overlooking factors such as differences in the quantity of RNA expressed by different cell types. Researchers at St. Jude created a more sophisticated algorithm that enables anyone performing RNA-seq analysis to efficiently account for these biological truths and potentially uncover new findings from existing data.

“ReDeconv incorporates the transcriptome size difference to improve RNA-seq analysis,” said corresponding author Jiyang Yu, Ph.D., St. Jude Department of Computational Biology interim chair. “Current methods make false assumptions that may negatively affect deconvolution, where we identify which cell types are in a sample, and other downstream analyses. We named the tool ReDeconv to suggest that researchers may have to re-analyze and deconvolute their RNA-seq data, as they may be able to reveal new information.”

Deconvoluting analysis assumptions

ReDeconv reduces problems that arise from the flawed mathematical assumptions of older tools. For example, different cell types have different total levels of RNA expression. However, scientists often treat each cell type’s transcriptome as if they have equivalent RNA amounts, despite the understanding that this is incorrect.

For example, red blood cells express one gene, hemoglobin, while a stem cell expresses 10,000 to 20,000 genes. This mismatch can overemphasize cell types with high total gene expression and underemphasize those with low expression in the bulk RNA-seq deconvolution. Similar issues arise from differences in gene length and variances in gene expression within a cell type, but the algorithm mitigates all three sources of error.

“When we added these biological parameters to our model, we significantly improved accuracy and precision,” said first author Songjian Lu, St. Jude Department of Computational Biology. “The reduction in error rate shows that ReDeconv has great potential to help people get more from their gene expression analysis by improving single-cell RNA-seq normalization and bulk deconvolution.”