Bulk RNA-Seq pipeline suggestion incorporating UMIs

Question

[steveh]

Hi,

I have 149 bulk RNA-Seq samples (100 bp, paired-end, Illumina) which have come from sequencing in the form of fastq triplets, i.e. pairs of reads plus a third fastq which contains only UMIs.

My first question is - do I need to use the UMIs at all or just ignore them?

So far I've ignored them, and used this workflow (on just 10 samples to begin with):

1. FastQC on raw reads
2. Align to full ref human genome using STAR (in: fastq, out: BAMs, sortedByCoord)
3. Produce counts using featureCounts
4. MultiQC on results produced so far
5. Analyse using DESeq2

This works, but ignores the UMIs completely. Results from multiQC show, from STAR:

and from featureCounts:

(note the fairly large percentages of unassigned multimapping reads there).

Alternatively, I've tried to incorporate the UMIs with this changed workflow:

1. FastQC on raw reads
2. Align to full ref human genome using STAR (in: fastq, out: BAMs, sortedByCoord)
3. Add the UMIs from the fastq files to the BAMs produced by STAR, using fgbio’s AnnotateBamWithUmis

but I'm getting lost down a rabit-hole now, adding more and more steps to this pipeline just in order to satisfy various errors I'm getting from downstream tools, e.g.

1. fgbio SortBam
2. fgbio SetMateInformation
3. fgbio GroupReadsByUmi
4. fgbio CallMolecularConsensusReads
5. samtools rehead, to add SM tag to BAMs
6. fgbio FilterConsensusReads (results in vastly reduced BAM file sizes)

for the moment I've stopped here - maybe I can use these BAM files, but this workflow is starting to feel over-complicated and I don't have confidence it's the correct way to go.

So to summarise:

• Do I need to incorporate the UMIs at all?
• If so, could anybody suggest a workflow?

Many thanks, Steve

[MatthewP]

Can not see images,.

[steveh]

apologies, corrected now

[GenoMax]

Have you tried to de-duplicate reads using UMI's alone or in combination with read alignment starts using umi_tools?

Answer 1

[i.sudbery]

Here is what I would recommend with umi-tools.

Extract the UMIs from the fastqs before mapping. You'll need to do this once for each of the non-UMI reads.

umi_tools extract --bc-pattern=NNNNNNNNNN -I umi_reads.fastq.gz --read2s-in=reads_R1.fastq.gz --read2-stdout | gzip > reads_R1.extracted.fastq.gz
umi_tools extract --bc-pattern=NNNNNNNNNN -I umi_reads.fastq.gz --read2s-in=reads_R2.fastq.gz --read2-stdout | gzip > reads_R2.extracted.fastq.gz

where the number of Ns in the bc-pattern matches the number of bases in the UMI.

You can then proceed to map these reads using STAR as before.

Once the reads are mapped, sorted and indexed, deduplicate the BAMs with umi_tools dedup:

umi_tools dedup -I mapped_reads.bam -S deduplicated_reads.bam --paired

Now you can proceed to quantify with featureCounts and analyse with Deseq2 as before.

[steveh]

Thanks Ian - would that be sorted by coordinate? (asking because the fgbio workflow seems to require re-sorting by Queryname)

[i.sudbery]

Yes, sorted by coordinate.

[steveh]

thanks - and for the dedup step, do I need the --paired option or is that assumed?

[i.sudbery]

Ooops. Yes, you will need the paired option, I'll edit the post.

[steveh]

great, thanks so much for taking the time to answer at this time of year!

[steveh]

Just to update after lots of testing - this is the method I settled on, although adding the UMIs to the already-aligned BAMs and then using umi_tools dedup also works fine.

I don't recommend the method mentioned in my original post, using fgbio.

[GenoMax]

Not to be pedantic but "this" meaning the method/answer suggested by @i.sudbery above? If so I can move that comment to an answer, which you can then accept to provide closure to this thread.

[steveh]

yes that's correct, the @i.sudbery answer. The general pointer to umi_tools is also useful, but Ian's answer is very specific.

[GenoMax]

You are able to accept more than one answer. Ian's comment has been moved to an answer now.

Answer 2

[swbarnes2]

Have you looked at umi-tools?

https://github.com/CGATOxford/UMI-tools

Answer 3

[padwalmk]

Hi, Check out the number of the read in UMI, If it's less than 1 or 2 % of total reads then you do not have to worry et al. But if it's more than 10 % than you have to do something about it.