search

CityUHK-CompBio / DeepCC

DeepCC: a novel deep learning-based framework for cancer molecular subtype classification
https://CityUHK-CompBio.github.io/DeepCC/
MIT License
20 stars 16 forks source link

Extracting information from DeepFeatures #9

Closed jebard closed 4 years ago

jebard commented 4 years ago

How do you link the DeepFeatures to the original msigDB? I have 10 DeepCC Features (feature 4 and 7 look interesting) but I'm unsure how to interpret what this means in a biological sense. Do you have any recommendations for next steps in the analysis?

gaofeng21cn commented 4 years ago

Hi Jonathan,

I think there are two ways. The first one is simple. We can calculate the correlation between Functional Spectra and Deep Features, then you can order the results by absolute correlation values. It will be very clear that which are highly correlated pathways from the name.

A fancy way to do the same thing is generating the activation map of the deep network, kind of visualisation of a DL network to show which input can activate the node of certain deep feature but we are still working on it to make it user friendly.

jebard commented 4 years ago

Perfect, thank you!

I did a simple correlation like this which returned correlations of each deep feature to each spectra.

c <- cor(deepFeaturesMatrix, functionalSpectraMatrix, method="pearson")