Robust adjustment of sequence tag abundance

The majority of next-generation sequencing technologies effectively sample small amounts of DNA or RNA that are amplified (i.e. copied) before sequencing. The amplification process is not perfect, leading to extreme bias in sequenced read counts. We present a novel procedure to account for amplification bias and demonstrate its effectiveness in mitigating gene length dependence when estimating true gene expression.

We tested the proposed method on simulated and real data. Simulations indicated that our method captures true gene expression more effectively than classic censoring-based approaches and leads to power gains in differential expression testing, particularly for shorter genes with high transcription rates. We applied our method to an unreplicated Arabidopsis RNA-seq dataset resulting in disparate gene ontologies arising from gene set enrichment analyses.

R code to perform the RASTA procedures is freely available on the web at www.stat.purdue.edu/∼doerge/.