Hello,
I have 107 RNAseq (1 is bad from 108 samples) samples for analysis using limma. We start from one line of stem cell, then clone it to 3 replicates lines(represent by 1,2,3), and each replicate we did 2 sister replicates (represent by A,B)again, so totally we have 6 subjects cell line. for each subject we did treatment experiment at 3 Passage time (represent by P8,P23,P32),at each Passage time and each subject, we did 6 treatment experiment with control (represent by T1,T2,T3,T4 and control1, control2).  T1,T2,T3,T4  use different medicine concentration increasingly. control1, control2 are without medicine. 

the target is :

My question is 

1. whether I can calculate coefficient within each block(subject)

2. how to estimate which subject is more sensitive to passage and treatment?

Thanks

To answer your first question, you need to set up a design matrix with subject-specific coefficients. You haven't described how you intend to model the experimental factors, but I would probably do something like this:

design <- model.matrix(~0 + Passage:Subject + Treatment:Subject)

Looking at colnames(design) gives:

 [1] "PassageP23:Subject1A"   "PassageP32:Subject1A"   "PassageP8:Subject1A"   
 [4] "PassageP23:Subject1B"   "PassageP32:Subject1B"   "PassageP8:Subject1B"   
 [7] "PassageP23:Subject2A"   "PassageP32:Subject2A"   "PassageP8:Subject2A"   
[10] "PassageP23:Subject2B"   "PassageP32:Subject2B"   "PassageP8:Subject2B"   
[13] "PassageP23:Subject3A"   "PassageP32:Subject3A"   "PassageP8:Subject3A"   
[16] "PassageP23:Subject3B"   "PassageP32:Subject3B"   "PassageP8:Subject3B"   
[19] "Subject1A:TreatmentTr1" "Subject1B:TreatmentTr1" "Subject2A:TreatmentTr1"
[22] "Subject2B:TreatmentTr1" "Subject3A:TreatmentTr1" "Subject3B:TreatmentTr1"
[25] "Subject1A:TreatmentTr2" "Subject1B:TreatmentTr2" "Subject2A:TreatmentTr2"
[28] "Subject2B:TreatmentTr2" "Subject3A:TreatmentTr2" "Subject3B:TreatmentTr2"
[31] "Subject1A:TreatmentTr3" "Subject1B:TreatmentTr3" "Subject2A:TreatmentTr3"
[34] "Subject2B:TreatmentTr3" "Subject3A:TreatmentTr3" "Subject3B:TreatmentTr3"
[37] "Subject1A:TreatmentTr4" "Subject1B:TreatmentTr4" "Subject2A:TreatmentTr4"
[40] "Subject2B:TreatmentTr4" "Subject3A:TreatmentTr4" "Subject3B:TreatmentTr4"

The first 18 coefficients represent expression of the control samples for each subject/passage combination. The remaining coefficients represent the log-fold change of the named treatment over the control for each subject. (We assume that, for each subject, the treatment effect is the same for each passage. You don't really have enough residual d.f. to allow for passage/treatment-specific treatment effects within each subject.) In this manner, you can compute coefficients for each level of the Subject blocking factor.

For your second question, I don't know what you mean by "sensitivity", but I'll assume that you want to find which subject has more DE genes across passages or across treatments. To find DE across passages for, say, subject 1A, you could do something like the below (replace colons with underscores in colnames(design), otherwise makeContrasts will complain):

con <- makeContrasts(PassageP23_Subject1A - PassageP32_Subject1A,
   PassageP8_Subject1A - PassageP32_Subject1A, levels=design)

This will test for DE between any of the passages for subject 1A. To test for DE across treatments, you would simply do:

con <- makeContrasts(Subject1A_TreatmentTr1, Subject1A_TreatmentTr2,
    Subject1A_TreatmentTr3, Subject1A_TreatmentTr4, levels=design)

... which will test if any of the treatments have a non-zero effect relative to the control for subject 1A. If you repeat this process for all of the subjects, you can figure out which subject is most "sensitive" based on which one has the most DE genes.