Add three functions to facilitate the use of nlme

1 files changed, 134 insertions, 0 deletions

diff --git a/man/nlme.Rd b/man/nlme.Rd
new file mode 100644
index 00000000..e74ba16d
--- /dev/null
+++ b/man/nlme.Rd
@@ -0,0 +1,134 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/nlme.R
+\name{mean_degparms}
+\alias{mean_degparms}
+\alias{nlme_data}
+\alias{nlme_function}
+\title{Estimation of parameter distributions from mmkin row objects}
+\usage{
+mean_degparms(object)
+
+nlme_data(object)
+
+nlme_function(object)
+}
+\arguments{
+\item{object}{An mmkin row object containing several fits of the same model to different datasets}
+}
+\value{
+A named vector containing mean values of the fitted degradation model parameters
+
+A groupedData data object
+
+A function that can be used with nlme
+}
+\description{
+This function sets up and attempts to fit a mixed effects model to
+an mmkin row object. An mmkin row object is essentially a list of mkinfit
+objects that have been obtained by fitting the same model to a list of
+datasets.
+}
+\examples{
+sampling_times = c(0, 1, 3, 7, 14, 28, 60, 90, 120)
+m_SFO <- mkinmod(parent = mkinsub("SFO"))
+d_SFO_1 <- mkinpredict(m_SFO,
+  c(k_parent_sink = 0.1),
+  c(parent = 98), sampling_times)
+d_SFO_1_long <- mkin_wide_to_long(d_SFO_1, time = "time")
+d_SFO_2 <- mkinpredict(m_SFO,
+  c(k_parent_sink = 0.05),
+  c(parent = 102), sampling_times)
+d_SFO_2_long <- mkin_wide_to_long(d_SFO_2, time = "time")
+d_SFO_3 <- mkinpredict(m_SFO,
+  c(k_parent_sink = 0.02),
+  c(parent = 103), sampling_times)
+d_SFO_3_long <- mkin_wide_to_long(d_SFO_3, time = "time")
+
+d1 <- add_err(d_SFO_1, function(value) 3, n = 1)
+d2 <- add_err(d_SFO_2, function(value) 2, n = 1)
+d3 <- add_err(d_SFO_3, function(value) 4, n = 1)
+ds <- c(d1 = d1, d2 = d2, d3 = d3)
+
+f <- mmkin("SFO", ds, cores = 1, quiet = TRUE)
+mean_dp <- mean_degparms(f)
+grouped_data <- nlme_data(f)
+nlme_f <- nlme_function(f)
+
+library(nlme)
+m_nlme <- nlme(value ~ nlme_f(name, time, parent_0, log_k_parent_sink),
+  data = grouped_data,
+  fixed = parent_0 + log_k_parent_sink ~ 1,
+  random = pdDiag(parent_0 + log_k_parent_sink ~ 1),
+  start = mean_dp)
+summary(m_nlme)
+
+\dontrun{
+  Test on some real data
+  ds_2 <- lapply(experimental_data_for_UBA_2019[6:10],
+   function(x) x$data[c("name", "time", "value")])
+  m_sfo_sfo <- mkinmod(parent = mkinsub("SFO", "A1"),
+    A1 = mkinsub("SFO"), use_of_ff = "min")
+  m_sfo_sfo_ff <- mkinmod(parent = mkinsub("SFO", "A1"),
+    A1 = mkinsub("SFO"), use_of_ff = "max")
+  m_fomc_sfo <- mkinmod(parent = mkinsub("FOMC", "A1"),
+    A1 = mkinsub("SFO"))
+  m_dfop_sfo <- mkinmod(parent = mkinsub("DFOP", "A1"),
+    A1 = mkinsub("SFO"))
+  m_sforb_sfo <- mkinmod(parent = mkinsub("SFORB", "A1"),
+    A1 = mkinsub("SFO"))
+
+  f_2 <- mmkin(list("SFO-SFO" = m_sfo_sfo,
+   "SFO-SFO-ff" = m_sfo_sfo_ff,
+   "FOMC-SFO" = m_fomc_sfo,
+   "DFOP-SFO" = m_dfop_sfo,
+   "SFORB-SFO" = m_sforb_sfo),
+    ds_2)
+
+  grouped_data_2 <- nlme_data(f_2["SFO-SFO", ])
+
+  mean_dp_sfo_sfo <- mean_degparms(f_2["SFO-SFO", ])
+  mean_dp_sfo_sfo_ff <- mean_degparms(f_2["SFO-SFO-ff", ])
+  mean_dp_fomc_sfo <- mean_degparms(f_2["FOMC-SFO", ])
+  mean_dp_dfop_sfo <- mean_degparms(f_2["DFOP-SFO", ])
+  mean_dp_sforb_sfo <- mean_degparms(f_2["SFORB-SFO", ])
+
+  nlme_f_sfo_sfo <- nlme_function(f_2["SFO-SFO", ])
+  nlme_f_sfo_sfo_ff <- nlme_function(f_2["SFO-SFO-ff", ])
+  nlme_f_fomc_sfo <- nlme_function(f_2["FOMC-SFO", ])
+
+  # Allowing for correlations between random effects leads to non-convergence
+  f_nlme_sfo_sfo <- nlme(value ~ nlme_f_sfo_sfo(name, time,
+       parent_0, log_k_parent_sink, log_k_parent_A1, log_k_A1_sink),
+     data = grouped_data_2,
+     fixed = parent_0 + log_k_parent_sink + log_k_parent_A1 + log_k_A1_sink ~ 1,
+     random = pdDiag(parent_0 + log_k_parent_sink + log_k_parent_A1 + log_k_A1_sink ~ 1),
+     start = mean_dp_sfo_sfo)
+
+  # The same model fitted with transformed formation fractions does not converge
+  f_nlme_sfo_sfo_ff <- nlme(value ~ nlme_f_sfo_sfo_ff(name, time,
+       parent_0, log_k_parent, log_k_A1, f_parent_ilr_1),
+     data = grouped_data_2,
+     fixed = parent_0 + log_k_parent + log_k_A1 + f_parent_ilr_1 ~ 1,
+     random = pdDiag(parent_0 + log_k_parent + log_k_A1 + f_parent_ilr_1 ~ 1),
+     start = mean_dp_sfo_sfo_ff)
+
+  # It does converge with this version of reduced random effects
+  f_nlme_sfo_sfo_ff <- nlme(value ~ nlme_f_sfo_sfo_ff(name, time,
+       parent_0, log_k_parent, log_k_A1, f_parent_ilr_1),
+     data = grouped_data_2,
+     fixed = parent_0 + log_k_parent + log_k_A1 + f_parent_ilr_1 ~ 1,
+     random = pdDiag(parent_0 + log_k_parent ~ 1),
+     start = mean_dp_sfo_sfo_ff)
+
+  f_nlme_fomc_sfo <- nlme(value ~ nlme_f_fomc_sfo(name, time,
+       parent_0, log_alpha, log_beta, log_k_A1, f_parent_ilr_1),
+     data = grouped_data_2,
+     fixed = parent_0 + log_alpha + log_beta + log_k_A1 + f_parent_ilr_1 ~ 1,
+     random = pdDiag(parent_0 + log_alpha + log_beta + log_k_A1 + f_parent_ilr_1 ~ 1),
+     start = mean_dp_fomc_sfo)
+
+  # DFOP-SFO and SFORB-SFO did not converge with full random effects
+
+  anova(f_nlme_fomc_sfo, f_nlme_sfo_sfo)
+}
+}