analysis method, single sample or multi samples

[suyanan]

Hi,

My experimental design consists of two groups with three replicates per group, totaling six samples.
For IsoQuant quantification, should I run each sample individually, run each group together, or run all samples together? Which approach would be better for downstream differential expression analysis?
Would analyzing multiple samples together affect the quantification results of individual samples?

Best wishes.

[andrewprzh]

Hi @suyanan

I'd suggest to run all samples together - it will create a single table with 6 columns. Counts are computed individually and do not affect each other. The only small detail - a spliced isoform needs at least one spliced read across all samples to be included in the table.

If you want to discover novel transcripts - I'd also suggest combining you samples together to get a single GTF.

Best
Andrey

[suyanan]

Hi,

Test IsoQuant 4.0.0 version installed by conda, command is

isoquant --threads 32 --data_type nanopore --reference UU_Cfam_GSD_1.0.fa --genedb UU_Cfam_GSD_1.0.gtf --complete_genedb --bam V1.UU_Cfam_GSD_1.0.minimap2.bam V2.UU_Cfam_GSD_1.0.minimap2.bam V3.UU_Cfam_GSD_1.0.minimap2.bam --labels V1 V2 V3 --output isoquantA --prefix A --force --sqanti_output

I test dog sampes (genome version UU_Cfam_GSD_1.0), compare single sample and multi samples, getting count results are different.
The differences are small in magnitude across hundreds of genes, but large for a small number of genes, with some single-sample counts being zero.

method	V1	V2	V3	gene
singleSample	0.00	349	0.00	ARHGDIA
multiSamples	325	349	173	ARHGDIA
singleSample	0.00	0.00	1491	RHOA
multiSamples	1731	1946	1491	RHOA

Upon visualization of the BAM file for one of these genes in IGV, we observed clear read coverage, indicating that the count should not be zero. This suggests that the single-sample count is erroneous, whereas the results derived from multi-sample analysis are biologically reasonable.
I check RHOA gene bellow, and LOC111091330 gene(in near position are) count is both zero in single-sample and multi-sample.

Best wishes.

[andrewprzh]

@suyanan

I see, could send me subset or read assignments produced by IsoQuant for these genes and different runs so I can try to investigate the case closer?