Question if DESeq can still be used after correction for T0?
1
0
Entering edit mode
Guest User ★ 13k
@guest-user-4897
Last seen 10.3 years ago
We use DESeq for our pooled shRNA screen analysis. Often this is about comparing a single cell-line for drug treatment (TR) vs untreated (UT) both after a certain number of days, fe. 7. In this experimental design the transfection on timepoint 0 (T0) is the same for both conditions, so that when comparing Treated/Untreated one doesnot have to correct for T0, as (TR7/T0) / (UT7/T0) = TR7/UT7. However in case of comparing different cell-lines, the transfection is different, so one need to correct for T0, for example by dividing the counts of both conditions with the counts of T0. The question is if one can then still use DESeq for differential analysis, perhaps after certain transformations of the data like a log2 transformation of the ratios in order to get a (more) symmetrical distribution? Or what else would be the appropriate statistical test? Thank your for your reply in advance! -- output of sessionInfo(): - -- Sent via the guest posting facility at bioconductor.org.
DESeq DESeq • 955 views
ADD COMMENT
0
Entering edit mode
Guest User ★ 13k
@guest-user-4897
Last seen 10.3 years ago
We use DESeq for our pooled shRNA screen analysis. Often this is about comparing a single cell-line for drug treatment (TR) vs untreated (UT) both after a certain number of days, fe. 7. In this experimental design the transfection on timepoint 0 (T0) is the same for both conditions, so that when comparing Treated/Untreated one doesnot have to correct for T0, as (TR7/T0) / (UT7/T0) = TR7/UT7. However in case of comparing different cell-lines, the transfection is different, so one need to correct for T0, for example by dividing the counts of both conditions with the counts of T0. The question is if one can then still use DESeq for differential analysis, perhaps after certain transformations of the data like a log2 transformation of the ratios in order to get a (more) symmetrical distribution? Or what else would be the appropriate statistical test? Thank your for your reply in advance! -- output of sessionInfo(): - -- Sent via the guest posting facility at bioconductor.org.
ADD COMMENT
0
Entering edit mode
Hi On 10/06/13 14:40, Cor Lieftink [guest] wrote: > We use DESeq for our pooled shRNA screen analysis. Often this is > about comparing a single cell-line for drug treatment (TR) vs > untreated (UT) both after a certain number of days, fe. 7. In this > experimental design the transfection on timepoint 0 (T0) is the same > for both conditions, so that when comparing Treated/Untreated one > doesnot have to correct for T0, as (TR7/T0) / (UT7/T0) = TR7/UT7. > However in case of comparing different cell-lines, the transfection > is different, so one need to correct for T0, for example by dividing > the counts of both conditions with the counts of T0. The question is > if one can then still use DESeq for differential analysis, perhaps > after certain transformations of the data like a log2 transformation > of the ratios in order to get a (more) symmetrical distribution? Or > what else would be the appropriate statistical test? No and Yes. No: Methods meant for count data must not be used with data which does not consist of counts, such as transformed or normalized data. Yes: Of course you can analyse such data as long as you refrain from performing any divisions and leave it to DESeq to do this. This is what generalized linear models (GLMs) are good for. You set up a data frame with one row for each sample and two factors as columns, say "time" (with levels "T7" and "T0") and "drug" (with levels "treated" and "untreated"). Then you fit two models (see the section on GLMs in the vignette): reduced model: count ~ drug + time full model: count ~ drug * time and compare them, as shown in the vignette. Simon
ADD REPLY
0
Entering edit mode
Cor, Sorry, this may not have been the answer to your question; let me try again. If I understand you right, your set-up is this: You have two cell lines (A and B), and for each of which you have three types of samples, namely - base line, i.e. before any treatment or incubation (B0) - incubated for 7 days without drug (U7) - incubated for 7 days with drug (T7) Now, the effect of the drug on cell line A (after 7 days) is effect_on_A = (T7_A / B0_A) / (U7_A / B0_A) = T7_A / U7_A as your wrote in your mail. Similarily, the effect on B is effect_on_B = (T7_B / B0_B) / (U7_B / B0_B) = T7_B / U7_B What would be now the difference (or: ratio) between the effect on An and on B? effect_on_B / effect_on_A = (T7_B / U7_B) / (T7_A / U7_A) So, for this question, you still don't need the base-line samples. To find genes whose reaction to the drug is different between the two cell lines, you hence compare the models count ~ cellLine + treatment and count ~ cellLine * treatment where the factor "cellLine" has the two levels "A" and "B" and "treatment" has the two levels "T7" and "U7". So, maybe I am confused but I am now unsure whether one needs the base-line samples at all. Simon
ADD REPLY

Login before adding your answer.

Traffic: 759 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6