Question

Small number of genes, drop in sft fit curve at higher threshold powers. why?

0

Entering edit mode

4.7 years ago

RNAseqer ▴ 280

This may be a silly question with an obvious answer, but here goes:

I have noticed that when I take my microarray dataset and parse it into smaller chunks (fewer probes/genes per chunk) there is a pattern that appears in my soft fit curve when performing WGCNA. Specifically I get a drop off in the scale independence curve. Instead of plateauing at around .8 or .9, the curve seems to plateau then plunges at some higher soft thresholding power.

Could anyone explain to me whats going on?

Also, is this a sign of something problematic in the data or is this a known phenomenon that occurs whenever you are working with few probes/genes. I notice this happening any time I drop the number of probes below say 2500 or thereabouts. I have looked around but really haven't found a good explanation of this anywhere thus far.

Thanks!

sft wgcna scale-free • 818 views

ADD COMMENT • link updated 4.7 years ago by Kevin Blighe 88k • written 4.7 years ago by RNAseqer ▴ 280

1

Entering edit mode

The fact that the curve reach a plateau an then plunges at higher powers is caused by a drastric drop in the connectivity. This is quite common when you randomly subsample your dataset as the majority of your genes are supposed to have very low connectivity while only few are highly conencted (hub). This is a fundamental aspect of 'scale free' networks.

ADD REPLY • link 4.7 years ago by andres.firrincieli 3.8k

score 0 · Answer 1 · 2020-04-01

Hey again,

Why would you divide your data into smaller chunks and then re-perform WGCNA?

In a WGCNA network, everything is technically connected to everything else. The idea of soft thresholding is to provide a suitable 'weighting' to these connections (edges) such that noise is eliminated and strong signals are amplified. As you can probably imagine, the ideal soft thresholding value differs from dataset to dataset.

In your case, one certainly cannot conclude that there is something problematic with your data just going by the soft thresholding result. In any case, how was your data processed? - this is more important to understand. Also, which filtering did you do, i.e., for low expressed genes?

Kevin