Entering edit mode
Hello,
I would like to summon Michael Love here and anyone else that can contribute. I was doing some research on DESeq2 and reading the following topic of the tutorial: Group-specific condition effects, individuals nested within groups, I was wondering if I can use this strategy to join different datasets, then each dataset will be treated as a group, each group will have their distinct individuals, and each individual will have two conditions (in common among all datasets). Does the design described account for the probable batch effect?
Thank you in advance.
Michael Love Could I please ask if a similar method would work my set of samples? I have 255 samples which have been generated from 76 subjects (46 with a disease and 30 controls). The samples were collected from these subjects at two timepoints. And these samples were then treated or untreated. My main question was to see the differences between disease and controls when treated at either timepoint (less interested at differences between timepoints).
Initially, was going to use a simpler design ~ grouped factor (Treated/Untreated x Disease/Control x Timepoint A/B) but wondered if I could account, in the model, for differences between subjects that are not attributable to treatment, disease or timepoint. The design ~ subject + group failed I think because subject can only have a disease or be a control. ~60% subjects have samples across timepoints and almost always (>90%) have paired treated and untreated samples.
Could you please advise what would work within DESeq 2?
There is an example of how you can do this in the DESeq vignette.
An alternative would be to fit a linear mixed model using the
variancePartitionpackage, which will not require complete cases.Thank you very much. I had read that and wondered if I used the recommendation in DESeq vignette, I would lose the pairing between groups? eg. subject 1 who is a control would have a sample in Group (Treated_Control_TimepointA), (Untreated_Control_TimepointA), (Treated_Control_TimepointB), (Untreated_Control_TimepointB).
Would it help to collapse the grouped factor to individual factors and then follow the vignette, creating a new column for disease.subject? And then use a design ~ timepoint + disease + disease:subject.n + disease:treatment?
Otherwise, will try variancePartition / dream. Thank you so much for helping.
Yes, making a group factor simplifies things, and you will need a new column for disease subject to avoid having a design that is not full rank.