PROBC: PREDICTING HI-C INTERACTIONS/TADs FROM EPIGENETIC AND TRANSCRIPTION MODIFICATIONS

The code has two parts. First part is to estimate model parameters. Second part is to infer TADs over trained model.

There are two sample files in this directory for training: train_order1_width2_preloglinear_nonparam_lambda3800.0_iter100_dommetadata.txt train_order1_width2_preloglinear_nonparam_lambda3800.0_iter100_domparams.txt

Proposed method ProbC has two parts:

1- estParams.py: estimates the model parameters. It has the following input parameters:

markerpath: marker files for corresponding domain partitions. It is a combination of marker data of different chromosomes. Marker data of each chromosome is separated by blank.

The format is as follows: marker \t bin \t value ....

where each line defines the value of a marker at a given bin.

domainpath: domain file. The format is as follows: nodecount\t1000 start,end\m ....

where start and end are the beginning and end of domains~(including both of them). First line says that we have 1000 bins. It is a combination of domain partition of different chromosomes. Domain partition of each chromosome is separated by blank.

outprefix: prefix of output files. It generates two output files. Metadata file: which includes information about the procedure, and domparams file: which includes the estimated model parameters.

lambdaval: regularization lambda for smoothness

grlambdaval: group lambda value

width: width of each bin on affecting TADs. We always use 1.

prepromodel: model for preprocessing input marker data (such as log transformation).

itercount: Maximum number of iterations of the optimization procedure.

basecount: number of kernel bases.

cb, ci, ce: relative weights of boundaries, interior and external segments.

2- findDomain.py: estimates TADs by using model parameters and markers.

markerpath: marker file similar to above. However, it has markers of a single chromosome to be predicted.

parampath: model parameters estimated above.

outprefix: output filename prefix

nodecount: maximum number of bins. 1, \ldots, nodecount.

nooverlap: whether predicted domains may overlap with each other.

prepromodel: marker preprocessing