Is it feasible to use population-level SNP instead of individual sample's variant information?

[zhangjy859]

Hi,

I am trying to count allele specific expression from RNA-seq data, beacuse I am working with specific cell model, I am not have paired SNV infomation, is it feasible to use population-level SNP instead of individual sample's variant information?

Regrads.

Zhang

[gmcvicker]

Hi Zhang,
There's not really anyway to do this with the WASP. It is possible to obtain genotypes directly from ATAC-seq or RNA-seq, however, and those genotypes could be used by WASP. E.g. you could try to identify heterozygotes by running GATK haplotype caller on the RNA-seq reads. You could also check out the QuASAR tool by Harvey et al. 2015 (doi: 10.1093/bioinformatics/btu802).

Hope this helps.

[zhangjy859]

Hi @mcvicker Thank you so much for quick response and useful infomations, but I still have a bit confusing, In fact, I known WASP is a wide known tools and been widely accepted for a long time, I also noticed many works run WASP with common SNP database, for example, dbSNP or 1kg SNP panel (such as He et. al., 2022, https://doi.org/10.1038/s41398-022-01913-1 and Quach et. al., 2016, https://doi.org/10.1016/j.cell.2016.09.024). So despite my reservations about doing so, I was a little surprised when I learned from your answer that such a scheme is not feasible. Is there a mistake in my understanding? Regrads. Zhang

[John-Drake]

I am also interested in zhangyl859 question.

[gmcvicker]

You need to have individual genotypes for allele specific read counts and estimates of allelic imbalance to make sense. For example, if you assume that somebody has a heterozygous genotype (A/G), but they are in fact homozygous (A/A), the allelic counts will be highly biased (i.e., almost all overlapping reads will have the 'A' genotype, except for the small subset of reads with a sequencing error at that nucleotide).

[John-Drake]

I wonder if I am misinterpreting @zhangjy859 questions.

For the read-filtering pipeline, you need individual genotypes and genotype liklihood scores for snp2h5 to generate its output. In one of the examples in snp2h5 README.md, it uses 1000 genome VCF. I interpreted this as the "population-level SNP data" that @zhangjy859 was referring to in his question as well as the two papers he posted. Thus, it is rational to use the 1000 genome VCF files as a reference for individuals not included in the 1000 genome.

If I am correct that it is fine to use a population-reference for the read-filtering pipeline (which requires individual genotypes and genotype-likelihood scores), why is it necessary to have individual genotypes or genotype-likelihood for the read-filtering pipeline alone? From reading the manuscript, my assumption is that WASP will (1) find mapped reads that include a SNP, (2) generate "new" reads that include those SNPs and re-map, and (3) then decide whether to keep SNP-containing reads based on whether they map to the same location.

In my head, all you would need is information about the individual SNPs (e.g. position and alleles).

My desire is to generate alignment files without mapping bias... and then use GATK ASEREadCounter.

Thank you for your help!
-John

[gmcvicker]

You could use this approach. The risk is that there will be many rare/uncommon variants in the individual(s) you are looking at that will not be present in 1000G. A subset of these variants could cause biased mapping and false signals of ASE.