WGCNA: Problem with selecting soft threshold below R^2 0.8

Question

[emilymily77]

Hi, I'm new with WGCNA so I have trouble dealing with it. I have 1560 samples, but R2 cutoff is 0.8 and I see that none of the scale free topology model fit is above that. What soft threshold power should I select?

       Power SFT.R.sq   slope truncated.R.sq mean.k. median.k. max.k.
1      1  0.73200  2.3400          0.896   704.0    734.00  995.0
2      2  0.30700  0.5700          0.641   407.0    424.00  700.0
3      3  0.00023  0.0131          0.383   265.0    271.00  522.0
4      4  0.08980 -0.2580          0.504   185.0    186.00  408.0
5      5  0.22200 -0.4520          0.619   135.0    133.00  328.0
6      6  0.32800 -0.6270          0.685   103.0     99.50  269.0
7      7  0.41800 -0.7310          0.759    80.5     76.00  225.0
8      8  0.48800 -0.7980          0.823    64.4     59.70  191.0
9      9  0.53600 -0.8660          0.847    52.4     47.60  164.0
10    10  0.57100 -0.9300          0.869    43.4     38.90  142.0
11    12  0.62500 -0.9900          0.913    30.8     26.80  110.0
12    14  0.67400 -1.0300          0.940    22.8     18.90   87.1
13    16  0.71800 -1.0700          0.954    17.4     13.70   70.4
14    18  0.75200 -1.0900          0.956    13.7     10.10   57.9
15    20  0.77800 -1.1400          0.973    10.9      7.73   49.1

[1769mkc]

you can choose 20 as I see its saturated there ...

Answer 1

[LChart]

What does a PCA plot of your samples look like? Do you have outliers?

On one large dataset (>300 samples) long ago, I had precisely this problem, and I downsampled to 50 samples (randomly, several times) and found that the soft threshold behaved "normally" -- and generally looked like the truncated threshold for the full dataset, so I went with the truncated.R.sq >90 cut. You might check if your data behaves similarly.