Hi all,

I am running WGCNA to analyze my 51 samples and I have a problem when choosing the best threshold. 

Here are the code I used and the result result I got. Should I pick 2 as the soft threshold?

Any help is appreciated. Thank you very much!

powers = c(c(1:10), seq(from = 12, to=20, by=2))
sft = pickSoftThreshold(datExpr, powerVector = powers, verbose = 2)

  Power SFT.R.sq  slope truncated.R.sq mean.k. median.k. max.k.
1      1    0.927 -2.940          0.977    6840    6230.0  14500
2      2    0.975 -3.140          0.982    1770    1480.0   5660
3      3    0.844 -2.380          0.859     868     768.0   3280
4      4    0.732 -1.940          0.895     568     545.0   2300
5      5    0.656 -1.730          0.884     429     390.0   1770
6      6    0.602 -1.520          0.879     351     287.0   1440
7      7    0.572 -1.340          0.788     302     218.0   1210
8      8    0.531 -1.090          0.617     269     169.0   1040
9      9    0.671 -0.803          0.614     245     133.0    913
10    10    0.676 -0.706          0.584     227     107.0    832
11    12    0.642 -0.736          0.540     201      70.3    804
12    14    0.653 -0.729          0.555     184      48.5    784
13    16    0.660 -0.714          0.564     171      34.7    769
14    18    0.669 -0.704          0.574     162      25.6    756
15    20    0.670 -0.699          0.575     154      19.2    746

You don't say how many genes/transcripts and samples you have in you data set and what conditions this data some from, so it's hard to judge.

I personally would be wary of using a power 2 in a regular unsigned network because of the large mean/median connectivities you see in the output. I usually work with RNA-seq data sets of  less than 100 samples summarized to about 20k genes and usually pick the soft threshold such that median connectivity is "on the order of" 30-50, which would lead to a high power of 12. Needing such high power to achieve what I consider a reasonable connectivity would make me go back and double-check the data. But all this depends on the data set, it is definitely not set in stone.