I have three time courses RNA-seq data, each with 7 time points. Let's say: A0-A6, B0-B6 and C0-C6. Group A is the control. Group B and C are the treatment. Experiments were practiced in different chamber. A0, B0 and C0 should have the same expression pattern because there was no treatment. B1 and C1 should have the same expression pattern because at this time point they have the same treatment. 

I extracted the FPKM value for each sample and constructed the co-expression gene network. I found that for some of the modules. A0, B0 and C0 have distinct expression patterns. B1 and C1 also have distinct expression patterns. 

So I think probably there are con-founding factors I need to remove. What I think of is 1) A, B and C were in different chambers; 2) when constructing the RNA-seq library, they were constructed in different time or by different technician. 

I performed SVA on the normalized reads counts. The following figure is what I got. 

Can anyone help me about how to interpret this result? Thanks. 

Here are the SVA values:

> svseq$sv
              [,1]         [,2]
 [1,]  0.009192911  0.162527163
 [2,] -0.034741036  0.058043556
 [3,] -0.066650845 -0.029763446
 [4,] -0.025683410  0.150880452
 [5,] -0.077683242  0.054009933
 [6,] -0.059638939  0.103939447
 [7,] -0.083615631  0.020937213
 [8,] -0.100669391 -0.123441789
 [9,] -0.089096028 -0.022792381
[10,] -0.062262057  0.141618931
[11,] -0.085975657  0.074219155
[12,] -0.011600395  0.155127797
[13,] -0.098461690 -0.114897535
[14,] -0.111000288 -0.020475921
[15,] -0.074651099 -0.018269073
[16,] -0.050121049  0.156185826
[17,] -0.128375754 -0.147027183
[18,] -0.098535227  0.040917625
[19,]  0.252178279  0.090352075
[20,] -0.096786808 -0.035441980
[21,] -0.038204809  0.116394561
[22,] -0.039443197  0.281063431
[23,] -0.087854619 -0.060282970
[24,] -0.142030150 -0.206853067
[25,] -0.058833091  0.198775913
[26,]  0.065595632 -0.088154229
[27,] -0.128115269 -0.201734270
[28,]  0.238099387  0.011724816
[29,]  0.176003638 -0.417296898
[30,] -0.109827212 -0.052693317
[31,]  0.243649855  0.076270453
[32,] -0.072014462  0.121986598
[33,] -0.147361504 -0.358804920
[34,]  0.142148851  0.001610613
[35,] -0.094772360 -0.151431219
[36,] -0.119123568 -0.093710371
[37,]  0.058988911 -0.008983051
[38,]  0.227471848 -0.194994226
[39,] -0.095148054  0.021928728
[40,]  0.144954687  0.186759619
[41,]  0.262629004 -0.010099494
[42,] -0.085343108  0.121634219
[43,] -0.081679651  0.124965239
[44,]  0.237969582 -0.018586135
[45,] -0.126941139 -0.075164637
[46,] -0.090763584  0.052616016
[47,]  0.215154191 -0.215134477
[48,] -0.107707531 -0.033701055
[49,]  0.224487838  0.046286749
[50,]  0.230686747 -0.130762213
[51,] -0.106115474 -0.012841340
[52,]  0.223765049  0.124055869
[53,]  0.229475439  0.025124565
[54,] -0.095624522  0.123380634

You don't really interpret singular values. You could do a PCA plot of your data, then fit a linear model using just the singular values and then plot a PCA plot of the residuals to see if the singular values corrected for whatever batch effect you may have, but in your case it looks like that is hopeless.

This is because (if I understand your experiment), you have confounded the treatment and batch (and maybe batches, if you did the RNA-Seq separately as well). In other words, you can't say if any differences between treatments are due to the batch effect or to biological differences.

If you were doing some sort of linear modeling and comparing say the differences between time 0 and time 1, then the batch differences would drop out, and you could make comparisons between treated and control. There is an example of doing this in the edgeR user's guide. But you say something about co-expression, so it's not clear exactly what you are trying to do. If you are trying to do WGCNA or something, I think you will have problems.