abnormal eigenvectors
shiyi-pan opened this issue · 4 comments
Hi, I used find_compartments function to get eigenvectors and compartments, but I think the distribution of eigenvectors is abnormal,here is my eigenvectors:
EigVect1.pdf
The eigenvalues are negative at both ends of the chromosome and positive at the middle of the chromosome. Generally speaking, when the eigenvalue is negative, the corresponding compartment should be B compartment, and B compartment is heterochromatin with lower interaction and fewer genes. However, it is generally believed that both ends of the chromosome are usually euchromatin, and there are more genes at both ends of the chromosome.Therefore, I think there is a contradiction here. Could you help me explain the problem reasonably. Thank you very much.
Hi,
As you said the eigenvectors doesn't seem to show you the compartments.
First of all, the sign of the eigenvalue does not tell you which type of compartment is. The eigenvectors are used to identify the transitions between compartments. Then to decide if a compartment is A or B you have to use some biological measure. For example in the examples (http://3dgenomes.github.io/TADbit/tutorial/tutorial_8-Compartments_and_TADs_detection.html#compartments) we use the amount of GC content because it's known that active regions have a higher ratio, but you can use whatever measure you like if you have a track of activity.
As for your example, I can't tell without seeing the matrix but it could be that either your matrix is not dense enough (you need more reads) or that the first eigenvector is telling you another kind of division, for example between arms. In the latter you will have to look at the second eigenvector.
David
Thank you for your reply, David.
first, there are 47,696,249 valid reads after filter and my genome size is 960Mb. So do you think the matrix is dense enough or not? Second,is TADbit support other track like gene density ？
Thank you again , David.
Seems a good amount of reads to me. Have you checked the second eigenvector?
You can use any track you wish as a list of values and use the rich_in_A parameter in find_compartments. Just be sure the size of the list matches the size of the column/rows of the matrix.
Thank you for your reply, David. The second eigenvector seems normal. The A compartments distribute at the end of the chromosome and the B compartments distribute near the centromere . I will try to use gene density as the track.Thank you for your help.