Question

DESeq2 biological replicates and RIN values

0

Entering edit mode

pmoranlosada • 0

@pmoranlosada-14001

Last seen 6.5 years ago

Hi all,

I'm sure you have already talked about how to treat biological replicates in RNA-Seq differential expression but I can't find information about how to incorporate RIN values into the DE analysis.

I have iPSCs where some genes have been duplicated (Group - DUP), deleted (Group- DEL) and I have the control group - CTL. The iPSCs are coming from 9 different subjects (3 in each Del, DUP and CTL groups). Each subject has 2 clones (clone X and clone Y) taken for the RNAseq, and each clone was grown in parallel (in 2 different wells) to create relicas (a, b in last column) . Here is my study design:

Sample	RIN	Group	Patient	Clone	Replicate
101_c1a	8.6	CTL	Subjet1	Clone1	a
101_c1b	9	CTL	Subjet1	Clone1	b
101_c3a	9.5	CTL	Subjet1	Clone2	a
101_c3b	9.7	CTL	Subjet1	Clone2	b
11_c1a	6	CTL	Subjet2	Clone3	a
11_c1b	8	CTL	Subjet2	Clone3	b
11_c6a	7.7	CTL	Subjet2	Clone4	a
11_c6b	8	CTL	Subjet2	Clone4	b
61_c3a	8.7	CTL	Subject3	Clone5	a
61_c3b	8.9	CTL	Subject3	Clone5	b
61_c5a	9.8	CTL	Subject3	Clone6	a
61_c5b	9.7	CTL	Subject3	Clone6	b
16A_c11a	7.5	DEL	Subject4	Clone7	a
16A_c11b	7.1	DEL	Subject4	Clone7	b
16A_c12a	8.9	DEL	Subject4	Clone8	a
16A_c12b	9.2	DEL	Subject4	Clone8	b
16B_c10a	9.4	DEL	Subject5	Clone9	a
16B_c10b	9.4	DEL	Subject5	Clone9	b
16B_c4a	10	DEL	Subject5	Clone10	a
16B_c4b	9.4	DEL	Subject5	Clone10	b
16C_c2a	6.7	DEL	Subject6	Clone11	a
16C_c2b	7.2	DEL	Subject6	Clone11	b
16C_c4a	6	DEL	Subject6	Clone12	a
16C_c4b	5.5	DEL	Subject6	Clone12	b
16X_c1a	5	DUP	Subject7	Clone13	a
16X_c1b	7.3	DUP	Subject7	Clone13	b
16X_c3a	5.9	DUP	Subject7	Clone14	a
16X_c3b	6	DUP	Subject7	Clone14	b
16Y_c2a	5.9	DUP	Subject8	Clone15	a
16Y_c2b	6.2	DUP	Subject8	Clone15	b
16Y_c4a	5.7	DUP	Subject8	Clone16	a
16Y_c4b	5.1	DUP	Subject8	Clone16	b
16Z_c1a	5.7	DUP	Subject9	Clone17	a
16Z_c1b	7.7	DUP	Subject9	Clone17	b
16Z_c8a	5.7	DUP	Subject9	Clone18	a
16Z_c8b	5.5	DUP	Subject9	Clone18	b

I'm interested on the differential gene expression at the Group level (controlvsdup, control vs del and dup vs del). But, should I consider all samples from the same group as biological replicates? The design matrix should be like :

1. ddseq <- DESeqDataSetFromMatrix(countData = counts, colData = metadata, design = ~ Group )

Or should I consider possible differences between patients and clones?

2. ddseq <- DESeqDataSetFromMatrix(countData = counts, colData = metadata, design = ~ Group + Patient)

3. ddseq <- DESeqDataSetFromMatrix(countData = counts, colData = metadata, design = ~ Group + Patient+Clone

And how would you incorporate the RIN values into design matrix?

ddseq <- DESeqDataSetFromMatrix(countData = counts, colData = metadata, design = ~ Group + RIN ) ?

Thanks a lot for your help. Any hint will be more than appreciated : )

Best,

Patricia

deseq2 differential gene expression covariates rnaseq • 1.8k views

ADD COMMENT • link updated 7.2 years ago by Michael Love 43k • written 7.2 years ago by pmoranlosada • 0

score 0 · Answer 1 · 2017-09-22

You can't compare directly across group and control for the subject and clone differences using fixed effects (you can compare fold changes across group, if you had pre/post, and this is an example in the vignette, but that's a different case). You would have to use duplicateCorrelation in limma to do direct comparisons and control for subject and clone correlations.

I don't recommend people add RIN to the design off the bat. For one thing, if it is confounded with your biological groups, you have bigger problems and just adding to the design doesn't fix them. What is the correlation with RIN among CTRL vs DEL and RIN among CTRL vs DUP, it seems like it would be high? If you have any very bad samples, I would recommend detecting them using QC measures as shown in the vignette. If they are not obvious outliers in e.g. a PCA plot, you can include them in DESeq2, and they will just appear as additional within-group variation for some genes. If they were to show up as outliers for some genes, DESeq2 will automatically remove these problematic counts, increasing sensitivity for those genes. Another option is to use limma with sample weights.