This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Choosing a high soft thresholding power in WGCNA, is it reasonable?

Hello,

I am analyzing RNA-Seq experiments using WGCNA. I am performing a consensus analysis of two different datasets.

When choosing soft-thresholding power using networkType="signed hybrid", one dataset (about 40 samples) exceeds SFTI of 0.9 at 12, however, another dataset (about 25 samples) exceeds at 26.

I am thinking about going to pick 26 as this power can ensure the scale-free topology for both networks, however, the official Q&A stated that the reasonable power is "less than 15 for unsigned or signed hybrid networks, and less than 30 for signed networks", in which 26 is high for signed hybrid network I'm intending to construct. I think the high power needed for the latter dataset is caused by an interesting biological variable that I do not want to remove effect.

In this case, is it reasonable to choose high power like 26? Are there any disadvantages or points that should be taken care of, when choosing high power?

wgcna
• 9,976 views • link
updated by boczniak767 boczniak767 • written by aristelliger • aristelliger

Why to do you think that this high power is due to an interesting biological variable and not an uninteresting batch effect? How similar are the 25 samples for this 26 power dataset? Have you look at the PCA to see if there is any major factor or batch effect?

• 0 views • link
written by Lluís R. Lluís R.

Thank you for the reply, and I apologize for not mentioning the reason. I performed PCA on vst normalized count, and clincal conditions I am comparing closely clustered together separately, thus I think the high power needed is due to this variation.

• 0 views • link
written by aristelliger • aristelliger

Thank you for the reply. For the latter dataset, the mean connectivity is 40 at 0.8 SFTI and 1 at 0.9 SFTI.

• 0 views • link
written by aristelliger • aristelliger

If the power at 0.8 STFI is still too high, I would follow the table in the WGCNA faq

• 0 views • link
written by andres.firrincieli andres.firrincieli

Thank you for your suggestion. Could you please advise me why higher power is inappropriate?

• 0 views • link
written by aristelliger • aristelliger

Because at very high powers your hubs genes might start loosing connectivity. When I face your problem, I usually select the power according to the table reported in the WGCNA faq by also taking into account that the mean connectivity must be below 100.

• 0 views • link
written by andres.firrincieli andres.firrincieli

I apologize for the late reply, and thanks for your reasonable explanation. Could you please post this as answer so as to I can accept it.

• 0 views • link
written by aristelliger • aristelliger

1 answer

Because at very high powers your hubs genes might start loosing connectivity. When I face your problem, I usually select the power according to the table reported in the WGCNA faq by also taking into account that the mean connectivity must be below 100.

• 0 views • link
written by andres.firrincieli andres.firrincieli

Where do you get the mean connectivity cut-off of 100? I see in the FAQ: 'If the scale-free topology fit index fails to reach values above 0.8 for reasonable powers (less than 15 for unsigned or signed hybrid networks, and less than 30 for signed networks) and the mean connectivity remains relatively high (in the hundreds or above), chances are that the data exhibit a strong driver that makes a subset of the samples globally different from the rest.' This suggests that mean connectivity should be below the hundreds, but I'm interested where you specifically get a value of 100 from?

• 0 views • link
written by jodi.thomas • jodi.thomas

I guess is any mean connectivity value below three digit numbers. Am I wrong?

• 0 views • link
written by andres.firrincieli andres.firrincieli

Sorry for the slow reply. I'm not actually sure to be honest, though that sounds sensible. Here it is also suggested that mean connectivity should be below the hundreds, based on the FAQ. There's a lot of information out there on using the first, scale independence graph to pick the soft thresholding power (e.g. pick the first power to cross the threshold of 0.8 or 0.9). But there is little information on how one should interpret and use the 2nd, mean connectivity graph to also aid in picking the soft thresholding power. Comments from anyone with experience/knowledge on how to use the mean connectivity to aid picking the soft thresholding power would be greatly appreciated!

• 0 views • link
written by jodi.thomas • jodi.thomas

Hey Thomas,

check this post on bioconductor. Peter Langfelder is one of the WGCNA devs.

mean connectivity graph to also aid in picking the soft thresholding power.

At very high power (aristelliger example) the mean connectivity is always below 100. In other cases (check the link), you have the opposite problem, you reach a SFT at very low power with very high mean connectivity values. This is the worst case scenario because you ends up with very few large modules, with thousands of highly connected genes, that are very difficult to dissect.

• 0 views • link
written by andres.firrincieli andres.firrincieli

To clarify things for future readers, in this bioconductor post the WGCNA developer wrote:

I aim for mean connectivity around 100 or less (typically between 30-50); median connectivity usually ends up around 10.

• 0 views • link
written by boczniak767 boczniak767

Log in to answer this question.

More posts like this