Bias Correction in RNA-Seq Short-Read Counts Using Penalized Regression

RNA-Seq produces tens of millions of short reads. When mapped to the genome and/or to the reference transcripts, RNA-Seq data can be summarized by a very large number of short-read counts. Accurate transcript quantification, such as gene expression calculation, relies on proper correction of sequence bias in the RNA-Seq short-read counts. We use a linear model for the sequence bias, which is much more flexible than the popular Poisson model. We fit the model using a penalized regression method, which allows for a significant dimension reduction. The algorithm is scalable for modeling RNA-Seq data. We demonstrate the excellent performance of our proposed method by applying it to real examples. The methods are implemented in open-source code, which is available in the R package lmbc.