Using Specific Regions of Genome as Reference for Alignment Using STAR

Question

[hkarakurt]

Hello everyone, I have lots of transcriptome data (RNA-Seq and full transcript single cell RNA-Seq) and I need to align them to genome using STAR. But I have storage some problems.

I will focus on a group of genes in downstream analyses. Is it possible to align the reads to specific parts of genomes (gene group) while using STAR or creating a custom reference from genome fasta that includes only specific regions and use it as reference?

Thank you in advance

Answer 1

[GenoMax]

That is possible but the question is it appropriate to do so.

If your data comes from entire genome/transcriptome then using a reduced representation reference always leads to the chance that STAR will try to align things to a location they may not have originated from.

Answer 2

[Pierre Lindenbaum]

please, don't. See Exome Sequencing: Masking The Non-Genic Sequences ?

[hkarakurt]

So doing such a thing in the alignment step practically causes mis-aligned reads. I will try to align all reads to whole genome and extract the regions using a bed file. I believe this will not create false positives or mis-aligned reads (at least not as much as the previous scenario).

Thank you for your answers.

Answer 3

[Buffo]

If the problem is the storage capacity, you can filter the bam file to those regions of interest, see this post: Extract Reads From A Bam File That Fall Within A Given Region.

[ATpoint]

I would not do that. With a few genes alone you are not going to do a meaningful analysis, for example normalization and DE needs a fair amount of genes to be robust. Even more so on single-cell level for QC purposes. If storage is limited then do it file by file, get bam, then the count matrix for a single sample. Delete bam, next one. Eventually concat the matrices into a single one. Or use salmon for everything which produces counts directly.

[Buffo]

I wouldn't do that either:

Is it possible to align the reads to specific parts of genomes (gene group) while using STAR

I only proposed an alternative to the problem assuming that further analysis doesn't need information about other regions:

But I have storage some problems.

That might be a better solution (you should post it as an answer to the question, not to my answer):

If storage is limited then do it file by file, get bam, then the count matrix for a single sample. Delete bam, next one.