estimate_distance_parameter: Calculate distance penalty parameter

In cicero: Precict cis-co-accessibility from single-cell chromatin accessibility data

Description

Function to calculate distance penalty parameter (distance_parameter) for random genomic windows. Used to choose distance_parameter to pass to generate_cicero_models.

Usage

estimate_distance_parameter(
  cds,
  window = 5e+05,
  maxit = 100,
  s = 0.75,
  sample_num = 100,
  distance_constraint = 250000,
  distance_parameter_convergence = 1e-22,
  max_elements = 200,
  genomic_coords = cicero::human.hg19.genome,
  max_sample_windows = 500
)

Arguments

cds	A cicero CDS object generated using make_cicero_cds.
window	Size of the genomic window to query, in base pairs.
maxit	Maximum number of iterations for distance_parameter estimation.
s	Power law value. See details for more information.
sample_num	Number of random windows to calculate distance_parameter for.
distance_constraint	Maximum distance of expected connections. Must be smaller than window.
distance_parameter_convergence	Convergence step size for distance_parameter calculation.
max_elements	Maximum number of elements per window allowed. Prevents very large models from slowing performance.
genomic_coords	Either a data frame or a path (character) to a file with chromosome lengths. The file should have two columns, the first is the chromosome name (ex. "chr1") and the second is the chromosome length in base pairs. See data(human.hg19.genome) for an example. If a file, should be tab-separated and without header.
max_sample_windows	Maximum number of random windows to screen to find sample_num windows for distance calculation. Default 500.

Details

The purpose of this function is to calculate the distance scaling parameter used to adjust the distance-based penalty function used in Cicero's model calculation. The scaling parameter, in combination with the power law value s determines the distance-based penalty.

This function chooses random windows of the genome and calculates a distance_parameter. The function returns a vector of values calculated on these random windows. We recommend using the mean value of this vector moving forward with Cicero analysis.

The function works by finding the minimum distance scaling parameter such that no more than 5 distance_constraint have non-zero entries after graphical lasso regularization and such that fewer than 80 nonzero.

If the chosen random window has fewer than 2 or greater than max_elements sites, the window is skipped. In addition, the random window will be skipped if there are insufficient long-range comparisons (see below) to be made. The max_elements parameter exist to prevent very dense windows from slowing the calculation. If you expect that your data may regularly have this many sites in a window, you will need to raise this parameter.

Calculating the distance_parameter in a sample window requires peaks in that window that are at a distance greater than the distance_constraint parameter. If there are not enough examples at high distance have been found, the function will return the warning "Warning: could not calculate sample_num distance_parameters - see documentation details".When looking for sample_num example windows, the function will search max_sample_windows windows. By default this is set at 500, which should be well beyond the 100 windows that need to be found. However, in very sparse datasets, increasing max_sample_windows may help avoid the above warning. Increasing max_sample_windows may slow performance in sparse datasets. If you are still not able to get enough example windows, even with a large max_sample_windows paramter, this may mean your window parameter needs to be larger or your distance_constraint parameter needs to be smaller. A less likely possibility is that your max_elements parameter needs to be larger. This would occur if your data is particularly dense.

The parameter s is a constant that captures the power-law distribution of contact frequencies between different locations in the genome as a function of their linear distance. For a complete discussion of the various polymer models of DNA packed into the nucleus and of justifiable values for s, we refer readers to (Dekker et al., 2013) for a discussion of justifiable values for s. We use a value of 0.75 by default in Cicero, which corresponds to the “tension globule” polymer model of DNA (Sanborn et al., 2015). This parameter must be the same as the s parameter for generate_cicero_models.

Further details are available in the publication that accompanies this package. Run citation("cicero") for publication details.

Value

A list of results of length sample_num. List members are numeric distance_parameter values.

References

• Dekker, J., Marti-Renom, M.A., and Mirny, L.A. (2013). Exploring the three-dimensional organization of genomes: interpreting chromatin interaction data. Nat. Rev. Genet. 14, 390–403.

• Sanborn, A.L., Rao, S.S.P., Huang, S.-C., Durand, N.C., Huntley, M.H., Jewett, A.I., Bochkov, I.D., Chinnappan, D., Cutkosky, A., Li, J., et al. (2015). Chromatin extrusion explains key features of loop and domain formation in wild-type and engineered genomes. Proc. Natl. Acad. Sci. U. S. A. 112, E6456–E6465.

See Also

generate_cicero_models

Examples

  data("cicero_data")
  data("human.hg19.genome")
  sample_genome <- subset(human.hg19.genome, V1 == "chr18")
  sample_genome$V2[1] <- 100000
  input_cds <- make_atac_cds(cicero_data, binarize = TRUE)
  input_cds <- reduceDimension(input_cds, max_components = 2, num_dim=6,
                               reduction_method = 'tSNE',
                               norm_method = "none")
  tsne_coords <- t(reducedDimA(input_cds))
  row.names(tsne_coords) <- row.names(pData(input_cds))
  cicero_cds <- make_cicero_cds(input_cds, reduced_coordinates = tsne_coords)
  distance_parameters <- estimate_distance_parameter(cicero_cds,
                                                     sample_num=5,
                                                     genomic_coords = sample_genome)

cicero documentation built on Dec. 10, 2020, 2 a.m.