Question about the output of hicexplorer hicPCA

Question

[jiqianzhaonotold]

To compute the A / B compartments , I used hicPCA and get pca1.bedgraph 、pca2.bedgraph . I'm very confused about the two bedgraph files . For example , In one bedgraph there is chr1 520000 560000 0.0045540143 , but In the other there is chr1 520000 560000 -0.004626316597 . So how should I classify A / B compartments by opposite values ?

Thanks! Best Qianzhao

Answer 1

[Devon Ryan]

This was addressed on the mailing list by Vivek, please limit your cross-posts.

For those coming to this via google, in general pca1 is what you want, but for a variety of reasons occasionally pca2 is the one that actually represents the compartments.

[jiqianzhaonotold]

Thanks for your help ! I have used eigenvector ( from juicer ) and hicPCA for computing A/B compartment in same resolution , I'm very confused about the results of two methods . In some chromosomes , the results of eigenvector and hicPCA are similar , but in some other chromosomes , the results of two methods are opposite (one is positive , the other is negative) . Someone suggested that I could change +/- . Indeed, it would make them look similar , but I think +/- is correlating with A/B compartments so we should not change the +/- , and if we change the +/- , which is compartment A ? which is B ? Can you give me any suggestions? Thank you !

[Devon Ryan]

The sign in PCA is arbitrary, you'll get opposite signs randomly depending on how exactly the PCA method is implemented. I recall that attempts are made within HiCExplorer to flip the signs such that the A compartment is always positive, but that sort of thing won't work 100% of the time. As always, have a look at the data and flip the sign if appropriate.