fit_binding_isotherm: Fit binding isotherm

In benjbuch/summerrband: Analyse Gel-shift Assays

Description

Fit binding isotherm

Usage

fit_binding_isotherm(
  x,
  formula,
  degree = NULL,
  type = "macro",
  INDEX = NULL,
  limits_lower = c(-Inf, +Inf),
  limits_upper = c(-Inf, +Inf),
  limits_K_d = c(0, 1000),
  start_K_d = 10^c(-1, 4),
  correlation = c(ab = 2, ac = 2, bc = 2, abc = 6)
)

Arguments

x	A data frame to be evaluated.
formula	A formula using variables of x; given as RL ~ L0, where RL is the variable that describes the (normalized) concentration of the complex [RL]_0, and L0 the concentration of titrated species [L]_0.
degree	The order of the binding polynomial; if NULL it is inferred from the number of variables on the LHS of formula. Up to third order binding polynomials are supported.
type	Either "micro" or "macro" (default).
INDEX	A column variable that shall be used to split the data into at most two sets, for which both will share higher order K. You will get other estimates as .x and .y fits.
limits_lower	Upper and lower bounds for the paramters to be fitted.
limits_upper	Upper and lower bounds for the paramters to be fitted.
limits_K_d	Upper and lower bounds for the paramters to be fitted.
start_K_d	Upper and lower bounds of the K_d grid-start parameters (see nls.multstart::nls_multstart).
correlation	The correlation factors between the microscopic binding constants; must be a named vector with "ab", "bc", "ac", "abc" as names.

Details

The formula LHS should evaluate to the observed binding isotherm, i.e. the number of ligands bound at the given concentration (RHS). Column arithmetics are accepted and even preferred as the number of variables in the LHS will be used to determine the binding degree: bound_1 + 2 * bound_2 ~ conc indicates a second degree binding polynom.

If type == "macro", the binding isotherm is computed for a macroscopic (thermodynamic) process using gpf_macro.

If type == "micro", the intrinsic binding constants of the individual binding sites are computed temptatively using correlations that yield to harmonic means. (In any system this is non-sensical.) The vector must be strictly named with "ab", "bc", "ac", "abc" as appropriate. Missing values will be set to 0, i.e., the result will be the same as for type == "macro".

The grid start approach will take 540 points across all model estimates.

If INDEX is a column name, the data is internally split into (exactly) two groups, for which the higher order K are shared. Note that the grid start approach will take 1296 points and computation may take longer.

Up to third order binding polynoms can be computed.

Note that all K_d values are internally log-transformed and returned as such to allow for an equal search depth across all orders of magnitude using nls.multstart::nls_multstart.

Examples

library(dplyr)
assay_file <- system.file("extdata", "gel_03.txt", package = "summerrband")
assay_data <- iqtl_meta(assay_file, list(conc = c(2^seq(10, 0), 0)))

test3 <- assay_data %>%
  group_by(gel_id) %>%
  tidyr::pivot_wider(names_from = band_id, values_from = vol_frac,
  id_cols = c(conc, group_vars(.))) %>%
  model_cleanly_groupwise(fit_binding_isotherm, formula = band_1 + 2 * band_2 ~ conc,
                          newdata = data.frame(conc = 10^seq(-3, 3, length.out = 100)))

tidyr::unnest(test3, tidy)

library(ggplot2)
model_display(test3) + scale_x_log10()

# fitting with shared estimates
assay_data_2 <- assay_data %>%
  group_by(gel_id) %>%
  tidyr::pivot_wider(names_from = band_id, values_from = vol_frac,
  id_cols = c(conc, group_vars(.))) %>%
  mutate(index_col = rep(c("a", "b"), times = 6))

# these examples will take some computation time ...
#
# fit_binding_isotherm(assay_data_2, formula = band_1 + 2 * band_2 ~ conc, INDEX = index_col)
#
# test4 <- assay_data_2 %>%
#   model_cleanly_groupwise(fit_binding_isotherm, formula = band_1 + 2 * band_2 ~ conc,
#                           INDEX = index_col,
#                           newdata = data.frame(conc = 10^seq(-3, 3, length.out = 100)))

benjbuch/summerrband documentation built on Dec. 19, 2021, 8:43 a.m.