Downstream analysis from STAR Alignment at transcript level

Question

[bassanio]

Dear Team,

I have two groups of samples which have went through the following pipeline

RAW FASTQ > Trimmed Fastq >STAR (Genome) >Htseq-Count & Cufflinks

I was going through the Vignettes from IsoformSwitchAnalyzeR and I am unable to follow how to move forward

Under Isoform/transcript quantification Option A:

It's says to use quantification from Salmon/kalisto. In STAR aligner I use the genome sequence to align against but according to salmon and Kalisto it requires transcriptome fasta(cDNA) as reference. So how move forward? Is it as discussed in the salmon where we redo the alignment to cDNA?

Under Isoform/transcript quantification Option B:

I felt this is the most suitable method from the pipeline similar to the existing method I used. The confusion for me is on the step 4 . After running the Cuffmerge for creating the merged gtf for including novel transcript should I need to run Cuffdiff ? or Is it that we use the gtf file from cuffmerge and use salmon downstream? If so the above issue of different reference exist here also (genome fasta and cDNA fasta) secondly for using the salmon with the new gtf file which fasta I need to use

Answer 1

[Rob]

For option A, you also need to instruct STAR to produce a transcriptome-coordinate BAM file. This is easy to do; when you run STAR you should pass additional options --quantMode TranscriptomeSAM and ---quantTranscriptomeBan Singleend. See page 18 of the STAR manual for a description of these features. With the transcriptome BAM file, you can then quantify abundance at the isoform level using salmon.

[bassanio]

@Rob Thank you for the reply. I have both Transcriptome bam and genome bam. But as you see the salmon documentation salmon needs bam/sam aligned to the cDNA reference not the Chromosome(DNA) reference. My concern is with this

[Rob]

The transcriptome bam file contains the genomic alignments projected to the transcriptome annotation. You should be able to see this if you look at the header of the AlignedToTranscriptome.out.bam file. So long as those records (sequence names and lengths) match the transcriptome file you pass to salmon, everything should be in working order. In fact, this very pipeline STAR -- projected to transcriptome --> salmon is the default quantification pipeline in nf-core/rnaseq.

[bassanio]

in the nfcore the salmon has index with both genome and transcriptome fasta before running salmon quant where as in the same pipeline in star only uses genome.fasta. My issue is that I don't have the transcript_fasta for which I have the used the genome fasta

cat $transcript_fasta $genome_fasta > $gentrome

salmon \\
    index \\
    --threads $task.cpus \\
    -t $gentrome \\
    -d decoys.txt \\
    $args \\
    -i salmon

But thank you I will follow as what you have suggested in the nfcore. Still I am Concerned [:-)]

[Ram]

Why the double back-slashes?

[Rob]

That is the indexing rule, but if you look at the quantification rule, you'll see that the way salmon is invoked is different depending upon if the user is using "alignment mode" or if the user is providing the raw FASTQ files directly to salmon for it's builtin selective-alignment. For example, see here. Not to attempt to make an argument from authority here ;P, but I am an author and maintainer of salmon, and we use it ourselves frequently with the AlignedToTranscriptome.out.bam from STAR. It works well (you can read more in this paper we published a few years ago "Alignment and mapping methodology influence transcript abundance estimation".

Perhaps your concern is that you don't have the transcriptome FASTA file itself to pass to salmon? That should be fairly easy to obtain from the genome fasta and the GTF file you provided to STAR, and can be done using a tool like gffread or the rsem-prepare-reference tool that you can see used in nfcore here.