secastel
commented
7 years ago Hi Maik, I'm glad that you have found our publications and software useful! I would say that your use case is spot on, you are thinking about everything the right way. I have a few tips for you though that will answer your questions.
First off, population pre-phasing isn't necessary, but it does help substantially. This is because it allows phASER to combine data across variants that are not read-back phased using the population phasing. However, in the case of exome-seq data it may actually not make a huge difference, since it is more beneficial for array or WGS genotype calls. You are correct in that all of the population phasing servers that I know of run on HG19 (GRCH37) and have not yet moved over to hg38. I haven't myself moved over either so I can't be of help there.
For running phASER, in your case you have two options.
Option 1) Population phase your input VCF. Then run phASER in two steps. In the first step combine all of the data you have for each individual (exome-seq reads + all of the scRNA-seq reads). It would probably be easier to merge all of the scRNA-seq BAMs into a single file using e.g. samtools merge rather than provide phASER with hundreds of files as input. What this run will do is to produce the best phasing possible using all of the data available. Next, take the resulting outputted VCF from that run and use that to run phASER individually for each of the scRNA-seq libraries. You can use the outputs per library to generate the gene level haplotype counts.
Option 2) For this option you do not need to population phase the VCF. As before, create a merged BAM of all of the scRNA-seq libraries and provide these with the exome-seq BAM as inputs to phASER. It's kind of inefficient, but run this for every scRNA-seq library, and provide a third BAM, which will be the single scRNA-seq library you would like to generate haplotype counts for. Make sure to set --haplo_count_bam to whichever bam is the single scRNA-seq library.
With respect to your question about obtaining separate read counts for each BAM input. Unfortunately, at this point I don't have any plans to implement such a feature, since it can be addressed in these specific use cases as I have written above, but it may be something I will consider in future.
I have been meaning to put together a tutorial on how to best use phASER for variant phasing, similar to the tutorial on using it to generate ASE data, that would go over this kind of use case, but have not yet had time to do so. I hope what I have written above makes sense. Please let me know.
mce1
Dear Stephane,
With great interest I have read your recent publication in Genome Biology and seen phASER, which I would really like to use for my own work. So I was wondering if I could ask your advise on some points.
I'm trying to assess allele-specific gene expression in single-cell RNA-Seq data from human blood cells. From each individual, I have a fairly large number of single-cell transcriptomes (approximately 200-300). In addition, I have one exome per individual. My aim is to obtain the most accurate ASE counts possible from each individual single-cell transcriptome.
My strategy so far is:
1) Call variants from the exome. (I have used a bcbio pipeline for this [https://bcbio-nextgen.readthedocs.io/en/latest/contents/testing.html#exome-with-validation-against-reference-materials].)
2) Map the sc-RNA-Seq reads to the reference genome. (I have used STAR for the mapping and then WASP to eliminate mapping biases.)
3) Next, I would like to use phASER to obtain gene-level haplotypic read counts for each individual transcriptome using the variant file from step 1 and the bam file(s) generated in step 2.
In step 3, of course, I would like to use the reads from all the sc-transcriptomes of a given individual for the read backed phasing but obtain separate read counts for each individual single cell. Do I understand you correctly that this in principle can be achieved using
--bam <list of all the bam files> --haplo_count_bam 1to output haplotypic read counts for bam file 1 only, while using the information from all the bam files for phasing? If yes, is it possible to obtain separate read counts for multiple bam files from the same run (e.g. specifying--haplo_count_bam 1, 2, ...) or would I have to run the script repeatedly, once for each bam file?I also meant to ask another question: In your tutorial you emphasize that the variant file should be pre-phased using a method like population phasing. I was wondering how important that is in the above situation (i.e. with a vcf that has been constructed from a single exome and with a really large number of RNA-Seq reads that can be used for read backed phasing). Do you think pre-phasing would still improve the quality of allelic counts here? I'm asking this because the various servers for phasing and imputation all require GRCh37-based coordinates, while I would prefer to work with hg38. (I have noticed that the bed files you provide are also based on hg19 but I suppose these could be lifted over to hg38 pretty easily?) Are you aware of any resources for population-based (pre-)phasing based on hg38 coordinates?
Would be great if you could have a look at this and give me some feedback.
Thanks a lot in advance,
Maik