Train integrated CASTLE model

Source: R/castle_integrated.R

Estimates the parameters of the statistical model for the integrated version the CASTLE algorithm based on a set of training samples, assumed to have no mutant DNA present.

train_integrated_ddpcr_model(background_samples, abc_grid_resolution = 25)

Arguments

background_samples

data.frame with cancer negative training samples as rows. These are the samples used to train/estimate the parameters of the model. At least the following columns should be present:

  • WildtypeOnlyDroplets

  • MutantOnlyDroplets

  • DoubleNegativeDroplets

  • DoublePositiveDroplets

If data from QuataSoft is available, these can be imported using import_QS_files.

abc_grid_resolution

The resolution of the 3D-grid on which the integrated LR-test statistic is approximated. Default is 25 (equal to a grid of 25^3 = 15,625 points).

Value

List of parameter estimates for the integrated CASLTE model. This should be used as input for test_tumor_sample_integrated.

Details

In the test an integrated version of the Likelihood Ratio test statistic is used, and this is approximated using a Riemann sum on a 3D-grid. The parameter abc_grid_resolution controls this approximation as the sum is calculated on abc_grid_resolution^3 points. Higher values give better approximation but longer computation times.