Search
Question: problem for finding soft thresholding in WGCNA algorithm
0
gravatar for modarzi
5 weeks ago by
modarzi10
modarzi10 wrote:

Hi,

 I need guide for my problem in WGCNA. As has been performed in WGCNA tutorial, before network construction, we have to find best soft thresholding (sft). So, we need to design a red line that corresponds to using an R^2 cut-off of h. For this purpose, after plotting “Scale independence” we have to run below code:

>  abline(h=0.9,col="red")

In my dataset by h = 0.9, I can’t see any line in my “Scale independence” diagram but when I change h value to 0.8(h = 0.8) then I can see the red line in my diagram and also have intersect with one of my candidate value for power.

My question is that can I change h value from 0.9 to 0.8? If no, what should I do for solving problem?

I appreciate if you share your comment with me.

 

Best Regards,

Mohammad

 

1

The problem with the plot is that the values barely reach the 0.8 threshold, so when you plot the line at 0.9 it is out of the y axis of the plot. Why you don't reach the 0.9 I can't say, probably it means that the dataset is not homogeneous.

Maybe there is too much variability and too few samples to find a good scale free network. If you could explain from what experiment are you using data it might help to diagnose the problem.

ADD REPLYlink written 5 weeks ago by Lluís Revilla Sancho430

Dear Lluis Revilla

Hello,

Thanks for your answer. my data set is taranscriptome profiling of  50 samples with 56000 gene types. I downloaded from TCGA and its workflow.type  is "HTSeq - FPKM-UQ" . Now, I don't know what should I do? can I use, default value for power? As you know, WGCNA tutorial has been suggested default value for unsigned network equal 6 and for signed network equal 12. 

I appreciate if you share your comment with me.

Best Regards,

Mohammad

ADD REPLYlink written 5 weeks ago by modarzi10
1

All those 50 samples are from the same type of cancer? If not the lack of power might be due to this.

ADD REPLYlink written 5 weeks ago by Lluís Revilla Sancho430

Yes, all samples belong to the one cancer. If I use h=0.8, my power will be 5. can I change 0.9 to 0.8 or it is better to use default value? which on is better and reasonable?

 

 

ADD REPLYlink written 5 weeks ago by modarzi10
1

Both of them are ok, after all there is not any guarantee that the biological systems are scale free. However, you will need to explain why you choose 0.8. It might be enough saying that you couldn't reach better scale-free indicators with higher power.

ADD REPLYlink written 5 weeks ago by Lluís Revilla Sancho430

Thanks for your comment:

for better choosing power, I print the result of "pickSoftThreshold" as below:

   Power SFT.R.sq  slope     truncated.R.sq      mean.k. median.k. max.k.
1      1    0.689       12.600          0.608             5770.00  6.14e+03  10700
2      2    0.265       -2.080          0.870              1210.00  1.17e+03   3750
3      3    0.290       -0.804          0.979                340.00  2.92e+02   1680
4      4    0.785        -2.730          0.953               120.00  9.10e+01    877
5      5    0.822        -2.310          0.822                 51.10  3.21e+01    505
6      6    0.661        -2.470          0.596                26.10  1.23e+01    387
7      7    0.363        -3.680          0.274                15.70  5.15e+00    344
8      8    0.362         -3.340          0.274                10.80  2.28e+00    318
9      9    0.349          -3.010          0.266                 8.35  1.07e+00    302
10    10    0.314        -3.010          0.185                 6.97  5.28e-01    292
11    12    0.315         -2.370          0.237                  5.63  1.45e-01    279
12    14    0.271         -2.290          0.122                5.06  4.51e-02    273
13    16    0.265         -2.130          0.121                4.77  1.54e-02    269
14    18    0.299         -2.340          0.168                  4.61  5.59e-03    267
15    20    0.295         -2.230          0.167                   4.51  2.14e-03    265

As you see, the highest value of SFT.R.sq  belongs to power = 5. So, if you want to decide for this situation which options do you prefer? 5 or 6(default value)?  

plot by h = 0.8.pdf:

https://www.dropbox.com/s/81hh152fbuo54ny/plot%20by%20h%20%3D%200.8.pdf?dl=0

plot by h = 0.9.pdf

https://www.dropbox.com/s/y6aogj4use5gb8r/Plot-SFT-Scale%20independence-Mean%20Connectivity-h%3D0.9.pdf?dl=0

ADD REPLYlink modified 5 weeks ago • written 5 weeks ago by modarzi10

I would use 5 as power, because the SFT.R.sq but also the other metrics are better for 5 rather than for 6.

ADD REPLYlink written 5 weeks ago by Lluís Revilla Sancho430

Thanks for your comment.I do apologize for my frequent questions.

you said about other metrics addition SFT.R.sq. what metrcis are as your porpuse for using 5 instead 6?

and next question:I will use power as 5 for unsigned network. based on my "pickSoftThreshold" result, wich power is suitable for constructing signed network? 5 or 12(default value)?

ADD REPLYlink written 5 weeks ago by modarzi10

You need to read the documentation and the articles surrounding WGCNA, but all of them are provided by pickSoftThreshold: the mean,k, median.k...

For a signed network you need to do the same justify your decisions. I won't make the decision for you.

ADD REPLYlink written 5 weeks ago by Lluís Revilla Sancho430
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 417 users visited in the last hour