Search
Question: time series, block by patient, between and within comparisons limma-voom
0
8 weeks ago by
aec40
aec40 wrote:

Dear all,

I have a time course RNA-seq analysis with the following design: 5 time points, 15 patients, 2 types of patients.

the questions I would like to answer:

a) expression differences across time points irrespective of patient type (t1 vs t2, t1 vs t3, ...)

b) expression differences between the two types of patients for every time point or in general for any time point

should I follow limma's user guide section 9.6.1 or 9.6.2 ?

should I add duplicatecorrelation function to account for repeated measures (same patients each time point)?

should I combine the factors time_point + type_of_patient in one single factor?

thanks for the clarifications.

This is the information I have: patient (p1 to p15), time (t1 to t5), group (A,B)

sample patient time group
1 p1 t1

A

2 p1 t2 A
3 p1 t3 A
4 p1 t4 A
5 p1 t5 A
6 p2 t1 A
7 p2 t2 A
8 p2 t3 A
9 p2 t4 A
10 p2 t5 A
11 p3 t1 B
12 p3 t2 B
13 p3 t3 B
14 p3 t4 B
15 p3 t5 B
16 p4 t1 B
17 p4 t2 B
18 p4 t3 B
19 p4 t4 B
... .... ... ...

optional design formulas:

1) ~patient+timegroup

2) ~0+timegroup and then duplicatecorrelation to block for patient

3) ~patient+time+group

modified 8 weeks ago by Aaron Lun21k • written 8 weeks ago by aec40

A design matrix would be helpful. Or at least some code describing the available sample annotation.

sure, I will edit my post.

How many samples do you have in total? If I understand the question it seems that you have three patients for each time points which would be then divided into the 2 types of patients you have. Or it is something helse

I have 5 time points per patient, in total 15 patients. These 15 patients can be further divided in two groups for question b)

For question a) Do you want the answer irrespective of the group or not? (for me the only one that makes sense is the 3rd option. because then you can make the contrasts to answer any question you might have.

Yes, irrespective of the group.

1
8 weeks ago by
Aaron Lun21k
Cambridge, United Kingdom
Aaron Lun21k wrote:

It seems like the best strategy would be your approach 2 (i.e., a combined time-group factor with blocking on patient in duplicateCorrelation). This will allow you to answer all of your questions with a single design matrix. For your first question, just compare the averages of each time point across groups, and for your second question, you can compare groups within each time point or the averages of each group across all time points.