Tips for network prediction: WGCNA vs ARACNE (or MINET)

Question

[piyushjo]

Hi!

I am planning to do some network analysis using WGCNA and MINET (that also uses aracne algorithm). From my understanding WGCNA is pure correlation while aracne uses special algorithm (DPI) to find direct transcriptional relation. My thought was to first use WGCNA to find highly correlated genes and compare how many of them are predicted to be direct neighbors according to ARACNE to reduce false positive. Does anyone has any tips to share to improve network prediction?

Thanks!

Answer 1

[fuyingxue]

Hi, piyushjo! I have the similar idea with you. I am now doing some analysis on a gene expression data for a CNS disease. I am also going to do WGCNA and Aracne analysis on this dataset.

The difference is that I think this two method are relatively independent and they serve to different aims. WGCNa is mainly designed to find the gene clusters (modules) which perform similar function, while ARACNE intends to detect the regulatory effects of a TF on its target genes.

SO my opinion is that after you get the gene clusters by WGCNA, then you can find out how these gene clusters are regulated through some master regulators. You can also try another algorithm called viper, which is also based on ARACNE and designed to predict the regulator activity changes in different conditions.

I have also found a review paper related to this topic, which I think it's useful to help us better understand these methods. https://academic.oup.com/bib/article/19/4/575/2888441

[piyushjo]

Hi Fuyingxue,

Thanks for your input and sharing the review article.