aboutsummaryrefslogtreecommitdiff
path: root/R/mkinfit.R
blob: 3ca389902a36b044c84abd9a2f47b073d1bc315a (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
# $Id$

# Copyright (C) 2010-2013 Johannes Ranke
# Contact: jranke@uni-bremen.de
# The summary function is an adapted and extended version of summary.modFit
# from the FME package, v 1.1 by Soetart and Petzoldt, which was in turn
# inspired by summary.nls.lm

# This file is part of the R package mkin

# mkin is free software: you can redistribute it and/or modify it under the
# terms of the GNU General Public License as published by the Free Software
# Foundation, either version 3 of the License, or (at your option) any later
# version.

# This program is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more
# details.

# You should have received a copy of the GNU General Public License along with
# this program. If not, see <http://www.gnu.org/licenses/>
utils::globalVariables(c("name", "value"))

mkinfit <- function(mkinmod, observed,
  parms.ini = "auto",
  state.ini = c(100, rep(0, length(mkinmod$diffs) - 1)), 
  fixed_parms = NULL,
  fixed_initials = names(mkinmod$diffs)[-1],
  solution_type = "auto",
  plot = FALSE, quiet = FALSE,
  err = NULL, weight = "none", scaleVar = FALSE,
  atol = 1e-8, rtol = 1e-10, n.outtimes = 100,
  trace_parms = FALSE,
  ...)
{
  # Get the names of the state variables in the model
  mod_vars <- names(mkinmod$diffs)

  # Get the names of observed variables
  obs_vars = names(mkinmod$map)

  # Subset observed data with names of observed data in the model
  observed <- subset(observed, name %in% obs_vars)

  # Define starting values for parameters where not specified by the user
  if (parms.ini[[1]] == "auto") parms.ini = vector()
  defaultpar.names <- setdiff(mkinmod$parms, names(parms.ini))
  for (parmname in defaultpar.names) {
    # Default values for rate constants, depending on the parameterisation
    if (substr(parmname, 1, 2) == "k_") parms.ini[parmname] = 0.1 
    # Default values for rate constants for reversible binding
    if (grepl("free_bound$", parmname)) parms.ini[parmname] = 0.1 
    if (grepl("bound_free$", parmname)) parms.ini[parmname] = 0.02
    # Default values for formation fractions
    if (substr(parmname, 1, 2) == "f_") parms.ini[parmname] = 0.2
    # Default values for the FOMC, DFOP and HS models
    if (parmname == "alpha") parms.ini[parmname] = 1
    if (parmname == "beta") parms.ini[parmname] = 10
    if (parmname == "k1") parms.ini[parmname] = 0.1
    if (parmname == "k2") parms.ini[parmname] = 0.01
    if (parmname == "tb") parms.ini[parmname] = 5
    if (parmname == "g") parms.ini[parmname] = 0.5
  }

  # Name the inital state variable values if they are not named yet
  if(is.null(names(state.ini))) names(state.ini) <- mod_vars

  # Transform initial parameter values for fitting
  transparms.ini <- transform_odeparms(parms.ini, mod_vars)

  # Parameters to be optimised:
  # Kinetic parameters in parms.ini whose names are not in fixed_parms
  parms.fixed <- transparms.ini[fixed_parms]
  parms.optim <- transparms.ini[setdiff(names(transparms.ini), fixed_parms)]

  # Inital state variables in state.ini whose names are not in fixed_initials
  state.ini.fixed <- state.ini[fixed_initials]
  state.ini.optim <- state.ini[setdiff(names(state.ini), fixed_initials)]

  # Preserve names of state variables before renaming initial state variable parameters
  state.ini.optim.boxnames <- names(state.ini.optim)
  if(length(state.ini.optim) > 0) {
      names(state.ini.optim) <- paste(names(state.ini.optim), "0", sep="_")
  }

  # Decide if the solution of the model can be based on a simple analytical
  # formula, the spectral decomposition of the matrix (fundamental system)
  # or a numeric ode solver from the deSolve package
  if (!solution_type %in% c("auto", "analytical", "eigen", "deSolve"))
      stop("solution_type must be auto, analytical, eigen or de Solve")
  if (solution_type == "analytical" && length(mkinmod$map) > 1)
      stop("Analytical solution not implemented for models with metabolites.")
  if (solution_type == "eigen" && !is.matrix(mkinmod$coefmat))
      stop("Eigenvalue based solution not possible, coefficient matrix not present.")
  if (solution_type == "auto") {
    if (length(mkinmod$map) == 1) {
      solution_type = "analytical"
    } else {
      if (is.matrix(mkinmod$coefmat)) {
	solution_type = "eigen"
      } else {
	solution_type = "deSolve"
      }
    }
  }

  cost.old <- 1e100 # The first model cost should be smaller than this value
  calls <- 0 # Counter for number of model solutions
  out_predicted <- NA
  # Define the model cost function
  cost <- function(P)
  {
    assign("calls", calls+1, inherits=TRUE) # Increase the model solution counter

    # Trace parameter values if quiet is off
    if(trace_parms) cat(P, "\n")

    # Time points at which observed data are available
    # Make sure we include time 0, so initial values for state variables are for time 0
    outtimes = sort(unique(c(observed$time,
		      seq(min(observed$time), max(observed$time), length.out=n.outtimes))))

    if(length(state.ini.optim) > 0) {
      odeini <- c(P[1:length(state.ini.optim)], state.ini.fixed)
      names(odeini) <- c(state.ini.optim.boxnames, names(state.ini.fixed))
    } else odeini <- state.ini.fixed

    odeparms <- c(P[(length(state.ini.optim) + 1):length(P)], parms.fixed)

    parms <- backtransform_odeparms(odeparms, mod_vars)

    # Solve the system with current transformed parameter values
    out <- mkinpredict(mkinmod, parms, odeini, outtimes, solution_type = solution_type, atol = atol, rtol = rtol, ...)

    assign("out_predicted", out, inherits=TRUE)

    mC <- modCost(out, observed, y = "value",
      err = err, weight = weight, scaleVar = scaleVar)

    # Report and/or plot if the model is improved
    if (mC$model < cost.old) {
      if(!quiet) cat("Model cost at call ", calls, ": ", mC$model, "\n")

      # Plot the data and current model output if requested
      if(plot) {
        outtimes_plot = seq(min(observed$time), max(observed$time), length.out=100)

        out_plot <- mkinpredict(mkinmod, parms, odeini, outtimes_plot, 
          solution_type = solution_type, atol = atol, rtol = rtol, ...)

        plot(0, type="n", 
          xlim = range(observed$time), ylim = range(observed$value, na.rm=TRUE),
          xlab = "Time", ylab = "Observed")
        col_obs <- pch_obs <- 1:length(obs_vars)
        lty_obs <- rep(1, length(obs_vars))
        names(col_obs) <- names(pch_obs) <- names(lty_obs) <- obs_vars
        for (obs_var in obs_vars) {
          points(subset(observed, name == obs_var, c(time, value)), 
            pch = pch_obs[obs_var], col = col_obs[obs_var])
        }
        matlines(out_plot$time, out_plot[-1], col = col_obs, lty = lty_obs)
        legend("topright", inset=c(0.05, 0.05), legend=obs_vars, 
          col=col_obs, pch=pch_obs, lty=1:length(pch_obs))
      }
    
      assign("cost.old", mC$model, inherits=TRUE)
    }
    return(mC)
  }
  fit <- modFit(cost, c(state.ini.optim, parms.optim), ...)

  # We need to return some more data for summary and plotting
  fit$solution_type <- solution_type

  # We also need the model for summary and plotting
  fit$mkinmod <- mkinmod

  # We need data and predictions for summary and plotting
  fit$observed <- observed
  fit$obs_vars <- obs_vars
  fit$predicted <- mkin_wide_to_long(out_predicted, time = "time")

  # Collect initial parameter values in two dataframes
  fit$start <- data.frame(initial = c(state.ini.optim, 
		  backtransform_odeparms(parms.optim, mod_vars)))
  fit$start$type = c(rep("state", length(state.ini.optim)), rep("deparm", length(parms.optim)))
  fit$start$transformed = c(state.ini.optim, parms.optim)

  fit$fixed <- data.frame(
    value = c(state.ini.fixed, parms.fixed))
  fit$fixed$type = c(rep("state", length(state.ini.fixed)), rep("deparm", length(parms.fixed)))

  bparms.optim = backtransform_odeparms(fit$par, mod_vars)
  bparms.fixed = backtransform_odeparms(c(state.ini.fixed, parms.fixed), mod_vars)
  bparms.all = c(bparms.optim, bparms.fixed)

  # Collect observed, predicted and residuals
  data <- merge(fit$observed, fit$predicted, by = c("time", "name"))
  names(data) <- c("time", "variable", "observed", "predicted")
  data$residual <- data$observed - data$predicted
  data$variable <- ordered(data$variable, levels = obs_vars)
  fit$data <- data[order(data$variable, data$time), ]
  fit$atol <- atol
  fit$rtol <- rtol
  # Return all backtransformed parameters for summary
  fit$bparms.optim <- bparms.optim 
  fit$bparms.fixed <- bparms.fixed
  fit$bparms.ode <- bparms.all[mkinmod$parms] # Return ode parameters for further fitting
  fit$date <- date()

  class(fit) <- c("mkinfit", "modFit") 
  return(fit)
}

summary.mkinfit <- function(object, data = TRUE, distimes = TRUE, ...) {
  param  <- object$par
  pnames <- names(param)
  p      <- length(param)
  covar  <- try(solve(0.5*object$hessian), silent = TRUE)   # unscaled covariance
  if (!is.numeric(covar)) {
    message <- "Cannot estimate covariance; system is singular"
    warning(message)
    covar <- matrix(data = NA, nrow = p, ncol = p)
  } else message <- "ok"

  rownames(covar) <- colnames(covar) <- pnames
  rdf    <- object$df.residual
  resvar <- object$ssr / rdf
  se     <- sqrt(diag(covar) * resvar)
  names(se) <- pnames
  tval      <- param / se
  modVariance <- object$ssr / length(object$residuals)

  param <- cbind(param, se)
  dimnames(param) <- list(pnames, c("Estimate", "Std. Error"))

  bparam <- as.matrix(object$bparms.optim)
  dimnames(bparam) <- list(pnames, c("Estimate"))

  ans <- list(
    version = as.character(packageVersion("mkin")),
    Rversion = paste(R.version$major, R.version$minor, sep="."),
	  date.fit = object$date,
	  date.summary = date(),
	  use_of_ff = object$mkinmod$use_of_ff,
    residuals = object$residuals,
    residualVariance = resvar,
    sigma = sqrt(resvar),
    modVariance = modVariance,
    df = c(p, rdf), cov.unscaled = covar,
    cov.scaled = covar * resvar,
    info = object$info, niter = object$iterations,
    stopmess = message,
    par = param,
    bpar = bparam)

  ans$diffs <- object$mkinmod$diffs
  if(data) ans$data <- object$data
  ans$start <- object$start

  ans$fixed <- object$fixed

  ans$errmin <- mkinerrmin(object, alpha = 0.05)

  ans$bparms.ode <- object$bparms.ode
  ep <- endpoints(object)
  if (length(ep$ff) != 0)
    ans$ff <- ep$ff
  if(distimes) ans$distimes <- ep$distimes
  if(length(ep$SFORB) != 0) ans$SFORB <- ep$SFORB
  class(ans) <- c("summary.mkinfit", "summary.modFit") 
  return(ans)  
}

# Expanded from print.summary.modFit
print.summary.mkinfit <- function(x, digits = max(3, getOption("digits") - 3), ...) {
  cat("mkin version:   ", x$version, "\n")
  cat("R version:      ", x$Rversion, "\n")
  cat("Date of fit:    ", x$date.fit, "\n")
  cat("Date of summary:", x$date.summary, "\n")

  cat("\nEquations:\n")
  print(noquote(as.character(x[["diffs"]])))
  df  <- x$df
  rdf <- df[2]

  cat("\nStarting values for optimised parameters:\n")
  print(x$start)

  cat("\nFixed parameter values:\n")
  if(length(x$fixed$value) == 0) cat("None\n")
  else print(x$fixed)
  
  cat("\nOptimised, transformed parameters:\n")
  printCoefmat(x$par, digits = digits, ...)

  cat("\nBacktransformed parameters:\n")
  printCoefmat(x$bpar, digits = digits, ...)

  cat("\nResidual standard error:",
      format(signif(x$sigma, digits)), "on", rdf, "degrees of freedom\n")

  cat("\nChi2 error levels in percent:\n")
  x$errmin$err.min <- 100 * x$errmin$err.min
  print(x$errmin, digits=digits,...)

  printdistimes <- !is.null(x$distimes)
  if(printdistimes){
    cat("\nEstimated disappearance times:\n")
    print(x$distimes, digits=digits,...)
  }

  printff <- !is.null(x$ff)
  if(printff & x$use_of_ff == "min"){
    cat("\nEstimated formation fractions:\n")
    print(data.frame(ff = x$ff), digits=digits,...)
  }

  printSFORB <- !is.null(x$SFORB)
  if(printSFORB){
    cat("\nEstimated Eigenvalues of SFORB model(s):\n")
    print(x$SFORB, digits=digits,...)
  }

  printcor <- is.numeric(x$cov.unscaled)
  if (printcor){
    Corr <- cov2cor(x$cov.unscaled)
    rownames(Corr) <- colnames(Corr) <- rownames(x$par)
    cat("\nParameter correlation:\n")
    print(Corr, digits = digits, ...)
  }

  printdata <- !is.null(x$data)
  if (printdata){
    cat("\nData:\n")
    print(format(x$data, digits = digits, scientific = FALSE,...), row.names = FALSE)
  }

  invisible(x)
}

Contact - Imprint