Python algorithm for Hi-C TAD identification based on graph Laplacian method (also known as Fiedler vector method)
The original paper and corresponding matlab implementation
First of all, I recommend that you familiarize yourself with the attached notebook. Don't hesitate to ask questions and contribute feedback for future improvements.
Use the command below to import the class and all required dependencies
from TAD_Laplacian import *
Create a new class instance with:
tad = TAD_Laplacian()
Use a method fit to find TADs of input matrix H.
tad.fit(H, #parameters )
Let's consider the method parameters:
- precision::int - serve as Hi-C matrix data limiter. If the pixel value more than precision then it changes with value of precision else doesn't change. In case of precision value equals 0 data limiter is off.
- connected::bool - if True, changes all the k,k+1 and k,k-1 subdiagonal elements to 1, else remains them unchanged.
- norm_laplacian::bool - if True, use normalized laplacian matrix, else unnormalized one.
- toeplitz::bool - if True, applying Toeplitz normalization to the input matrix, else not.
- recursive::bool - if True, recursively apply the algorithm to subTADs of found TADs until the stop conditions will be achieved, else return only first iteration of Fiedler vector algorithm.
- minlimit::int - set minimum number of bins in a found TAD to continue of its decomposing recursively (it's the one of the stop conditions).
- threshold::float - maximum Fiedler value of a found TAD to stop decompose it recursively (it's the second stop condition)
After the previous method finished the work, you may visualize obtained result with next method:
tad.visuzalize()
It may be very useful to investigate results with some data transformation of the original matrix H. For such purpose there is a parameter transform_func which takes function (or lambda function). Note that after functional transformation of input matrix H, the out matrix must have the same shape.