About TIMP

TIMP implements partitioned variable projection for separable nonlinear least-squares problems. By splitting conditionally linear parameters from intrinsically nonlinear ones it reduces the effective search space and, more importantly, the memory footprint — making it practical to fit large time-wavelength matrices that would be expensive with standard variable projection.

The name refers back to the earlier tim collection of FORTRAN routines for modeling time-resolved spectroscopy data, which could not be published due to library licensing constraints. TIMP was the first attempt to free this software by rebuilding it on the open foundation of the R ecosystem as an R package — hence, presumably, the name.

Published in the Journal of Statistical Software in 2007, TIMP is the R engine on which the original Glotaran desktop application was built. The Java GUI added interactive data exploration and visual model specification, while TIMP remained the computational core.

TIMP is designed around what its authors call “interactive scientific model discovery”: formulate a candidate model, fit it, inspect the residuals and parameter estimates, then refine the model until the fit is satisfactory.

Capabilities

TIMP handles multiple datasets collected under related conditions, sharing parameters across experiments where appropriate and fitting others per dataset.

Kinetic models

Sequential and parallel compartment schemes
Full transfer-matrix (target) models
Multiexponential kinetics with any number of components

Instrument response and dispersion

Gaussian or measured instrument response functions
Wavelength-dependent dispersion modeling
Coherent artifact terms

Spectral parameterizations

Gaussian, Lorentzian, Voigt, and skewed-Gaussian lineshapes
Spline-based spectral descriptions

Constraints and parameter relations

Optional non-negative (or non-positive) least-squares estimation of conditionally linear parameters — the work on TIMP led to the standalone nnls CRAN package, an R interface to the Lawson–Hanson NNLS algorithm
Positivity and zero constraints on selected nonlinear parameters
Linked parameters across spectral regions or datasets
Shared and dataset-specific parameter partitioning

Additional domains

Fluorescence Lifetime Imaging Microscopy (FLIM)
Any data described by a superposition model with separable linear/nonlinear parameters

Representative Example

The code below fits two transient spectroscopy datasets measured at different laser intensities with a five-component sequential model, Gaussian IRF, wavelength-dependent dispersion, a coherent artifact term, and a mix of shared and dataset-specific parameters:

library(TIMP)

psi_1 <- preProcess(readData("psi_1.txt"), scalx2 = c(3.78, 643.5))
psi_2 <- preProcess(readData("psi_2.txt"), scalx2 = c(3.78, 643.5))

model <- initModel(
  mod_type = "kin",
  kinpar = c(7.9, 1.08, 0.129, 0.0225, 0.00156),
  irfpar = c(-0.1018, 0.0434),
  parmu = list(c(0.230)),
  lambdac = 650,
  seqmod = TRUE,
  positivepar = c("kinpar"),
  cohspec = list(type = "irf")
)

result <- fitModel(
  list(psi_1, psi_2),
  model,
  modeldiffs = list(
    dscal = list(list(to = 2, from = 1, value = 0.5))
  )
)

Compartments, rate constants, instrument response, dispersion, artifacts, spectral constraints, and dataset-to-dataset differences are all expressed within a single fit.

Where TIMP is used

TIMP is applied wherever multi-way data can be described as a superposition model with separable parameters:

Transient absorption spectroscopy
Time-resolved fluorescence
Fluorescence Lifetime Imaging Microscopy (FLIM)
Multi-experiment global and target analysis across related datasets

Installation

From CRAN:

install.packages("TIMP")

Or the development version from GitHub:

# install.packages("remotes")
remotes::install_github("glotaran/TIMP")

Citation

If you use TIMP, please cite:

Mullen KM, van Stokkum IHM (2007). TIMP: An R package for modeling multi-way spectroscopic measurements. Journal of Statistical Software 18(3).
DOI: 10.18637/jss.v018.i03