Dear Gordon and others,
I have a limma/design matrix related question that I was hoping to get some advice.
Our study contains a number of different treatment groups. We will call this parameter “gr” In each treatment group, we have a number of time points - the same time points for all individuals. We call this parameter “ti”. The first time point is the pre treatment baseline. Finally, the treatment produced one of two outcomes. We call this parameter “oc”
In summary, each individual received one treatment, placing the individual in one treatment group (“gr”). They are labelled A-E. From each individual, we obtained one baseline sample, and a time series during treatment (“ti"). They are labeled Day1, Day2… etc At the end of the study, each individual manifested one outcome (“oc”). They are labeled Favorable and Unfavorable.
We are interested in knowing what genes are differentially expressed by each treatment. We are also interested in what sets apart the individuals that received the favorable outcome.
We attempted to answer these questions in the following way:
We constructed a model matrix in the following way:
model.matrix(~tiocgr+ID)
ID represents the study individuals. We are trying to block them from influencing the analysis. When we do this, then inte Intercept term ends up including the baseline time point, groupA, with the Unfavorable outcome.
We find that setting up the contrasts becomes rather complex. We set them up as follows, with the corresponding motivations:
“tiDay1”, “tiDay2”,… etc - These contrasts show the response to treatment in groupA at each time point, compared to baseline
“tiDay1.ocFavorable-ocFavorable” - This contrast shows the response treatment in the protected individuals of groupA, compared to baseline, at a given time point.
“tiDay1.grB-grB”, “tiDay2.grB-grB”,… etc - These contrasts show the response to treatment in groupB at each time point, compared to baseline
“(tiDay1.grB.ocFavorable-grB.ocFavorable) - (tiDay1.grB-grB)” This contrast shows how the Favorable individuals responded to treatment at Day1, compared to how all the group in general responded to treatment at Day1.
Does this approach make sense, given the questions that we are trying to answer?
