musyc_fit_fn

Fit function for the MuSyC (multi-dimensional synergy of combinations) model on combination dose-response data, for use with apply_fit.

Usage

musyc_fit_fn(avg_dt, n_restarts = 10L, seed = 1L)

Arguments

avg_dt

data.table for one (drug1 x drug2 x cell line x normalization_type) combination. Must contain columns Concentration, Concentration_2, x, and normalization_type.

n_restarts

integer; number of restarts from random initial values.

seed

integer; seed for the random initial values. The global random number generator is left as it was.

Value

Named list with normalization_type, the parameters E0, E1, E2, E3, h1, h2, C1, C2, alpha12, alpha21, the efficacy synergy beta, r2 and converged. Numeric fields are NA_real_ when the fit does not converge.

Details

The surface is fitted with the Python synergy package through reticulate; install it with reticulate::py_install("synergy", pip = TRUE). The fit provenance records the Python, synergy, numpy and scipy versions.

The whole grid is fitted, single-agent edges included, with the synergy defaults and three additions: the untreated corner is set to 1 (added when absent), E is bounded from -1 for GR and from 0 for RV up to 1.5, and C1/C2 stay within a factor of 1000 of the tested doses. The fit starts from the synergy initial guess and from n_restarts random initial values; the fit with the best r2, computed on the measured points, is kept.

alpha > 1 means synergistic potency, beta > 0 synergistic efficacy. alpha is weakly constrained when a drug has little effect, so extreme values of it are not a measure of synergy.

See also

musyc_summary_fn

Examples

if (musyc_available()) {
  mae <- gDRutils::get_synthetic_data("finalMAE_combo_matrix_small")
  combo_se <- mae[[gDRutils::get_supported_experiments("combo")]]
  combo_se_out <- gDRutils::apply_fit(
    combo_se, musyc_fit_fn, "combination",
    output_assay = "musyc_params",
    summary_fn = musyc_summary_fn, summary_assay = "musyc_summary",
    fit_source = "musyc"
  )
}