eigenvector output file is empty

[Saumya]

Dear all,

I am trying to run eigenvector  command on 100kb resolution however, it is ends without output. I am already using "-p" flag  and able to get output at lower resolution.

java -Xmx50000m   -jar ./juicer_tools.2.20.00.jar eigenvector  -p VC  ./merged30.hic chr1 BP 100000

WARNING: sun.reflect.Reflection.getCallerClass is not supported. This will impact performance.
WARN [2024-04-15T20:17:46,201]  [Globals.java:138] [main]  Development mode is enabled

WARNING: Eigenvector calculation at high resolution can take a long time

Any advice from the forum will be very helpful,

Many thanks in advance,

Best,

Saumya

[Moshe Olshansky]

Hi Saumya,

You can try our R or C++ POSSUMM suite from https://github.com/moshe-olshansky/EigenVector

There you can do almost any resolution (if it is available in your HiC file). I would recommend using o/e (which is the default) with KR or SCALE normalization - whichever is available in your file with the requested resolution.

Best regards,

Moshe.

[Saumya]

Dear Moshe,

Sorry for late response and thank you for the reply.

Thanks for pointing to the software. I will run it on my data now.

Best,

SAumya

[Aditya Shrivastava]

Dear Moshe, 

I was trying to use the inbuilt eigenvector command in the juicer at a resolution greater than 250 Kb, but I was facing the same problem. So I used your script written in R. To check the consistency, I used your script and inbuilt script at 250 Kb. Here are a few datasets - 
Algorithm-1 is the inbuilt eigenvector in juicer and Algorithm-2 is the R script.

 

for some chromosomes, both algorithms give similar results (as shown above). However, for many, there is a significant difference, especially for VC normalisation (shown below); which one should be correct?




Regards
Aditya Shrivastava

[Moshe Olshansky]

Hi Aditya,

Please note that the sign of the eigenvector is arbitrary. What you need to check is whether the correlation between the two eigenvectors is very close to either 1 or -1. Please note that if calling an eigenvector not with juicer tools, it is a good idea to provide the length (in bp) of your chromosome/contig to ensure the correct number of entries (bins). Otherwise trailing extremely sparse bins will be omitted.

--
You received this message because you are subscribed to a topic in the Google Groups "3D Genomics" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/3d-genomics/QPGOOp34ODk/unsubscribe.
To unsubscribe from this group and all its topics, send an email to 3d-genomics...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/3d-genomics/f90dc09a-9682-4cd6-8612-d9ab1347bb79n%40googlegroups.com.

[Aditya Shrivastava]

Dear Moshe, 

Thank you for your prompt reply. I have another query regarding eigenvectors-
 
I'm trying to replicate a compartment analysis similar to the one performed in the 2024 Woolly Mammoth study ("Compartment analyses at 1-Mb resolution...") - 


Specifically, I want to understand how to calculate the similarity between eigenvectors derived from different species? Do I have to calculate the correlation coefficient of eigenvectors chromosome-wise, take the magnitude of that coefficient, and then compute a weighted mean or average to generate the similarity matrix and dendrogram?  Because as I understood in some cases they are totally inverted to each other, leading to a Pearson's correlation coefficient close to -1. Currently I am using  juicer tool.

Regards
Aditya Shrivastava

[Olga Dudchenko]

Hi Aditya,

Since the sign is arbitrary you usually want to flip (multiply by -1) eigenvectors if necessary. You can just do by correlating the two eigenvectors and choosing a random (but consistent) orientation for both. Or you can do something like assign based on genes/GC content, e.g. for the open one to be consistently associated with positive eigenvalues. We did the latter for the paper, but it would not matter if you are solely interested in correlation analysis.

Best,

Olga

[Aditya Shrivastava]

Dear Olga and Moshe,

Thank you for your guidance.

Regards
Aditya Shrivastava