We are trying to sort out a pipeline for doing allele specific expression. Our plan is to call SNPs from RNA-seq data and combine with known SNP annotations. A well known problem in ASE is reference bias, where reads are more like to map if they have the reference sequence than the alt sequence.

  1. We have come accross two possible ways to deal with this. In many papers (e.g. Zhuo et al, Sigurdsson et al) a new reference is generated, substitute N bases where there are SNPs using FastaAlternativeReferenceMaker from GATK and reads are remapped to that.

  2. HISAT2 offers the possibility to provide SNP calls to the index builder, and build an index that accounts for these SNPs.

Does anyone have any advice on which of these might be better?


1 Answer 1


As you said, this is a well-known problem for ASE. In Degner et al. 2009, they state that, "Perhaps surprisingly, we found that masking known SNPs does little to eliminate inherent biases in read-mapping."

As such, substituting N bases is not likely to completely solve your issue. Depending on the aligner, these N bases may also negatively affect your ability to map reads.

Of your two suggestions, I believe HISAT2 will be the better approach. There are also more intensive methods you can use such as empirical Bayesian methods or more traditional variant phasing (ex. AlleleSeq).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.