rBiasCorrection

rBiasCorrection is published in ‘BiasCorrector: fast and accurate correction of all types of experimental biases in quantitative DNA methylation data derived by different technologies’ (2021) in the International Journal of Cancer (DOI: https://onlinelibrary.wiley.com/doi/10.1002/ijc.33681).

rBiasCorrection is the R implementation with minor modifications of the algorithms described by Moskalev et al. in their research article ‘Correction of PCR-bias in quantitative DNA methylation studies by means of cubic polynomial regression’, published 2011 in Nucleic acids research, Oxford University Press (DOI: https://doi.org/10.1093/nar/gkr213).

Installation

CRAN version

You can install rBiasCorrection simply with via R’s install.packages interface:

install.packages("rBiasCorrection")

Development version

If you want to use the latest development version, you can install the github version of rBiasCorrection with:

install.packages("remotes")
remotes::install_github("kapsner/rBiasCorrection")

Example

This is a basic example which shows you how to correct PCR-bias in quantitative DNA methylation data:

library(rBiasCorrection)

# define input file paths
experimental <- file.path(tempdir(), "/experimental_data.csv")
calibration <- file.path(tempdir(), "/calibration_data.csv")

# create example files from provided example dataset
data.table::fwrite(
  rBiasCorrection::example.data_experimental$dat,
  experimental
)
data.table::fwrite(
  rBiasCorrection::example.data_calibration$dat,
  calibration
)

# run bias correction algorithm
biascorrection(
  experimental = experimental,
  calibration = calibration,
  samplelocusname = "BRAF"
)

More detailed information on how to use the package rBiasCorrection can be found in the vignette and the FAQs.

Available Fitting Options (TODO)

There are three fitting options available for fitting the non-linear least squares (nls) algorithm with rBiasCorrection. The default method (used in the publication) is to fit nls with the Gauss-Newton algorithm and define for each parameter that should be optimized a random grid between -1000 and 1000 for initializing the starting estimates (options(rBiasCorrection.nls_implementation = "GN.paper").
For making a better guess on the starting estimates when fitting nls with the Gauss-Newton algorithm (options(rBiasCorrection.nls_implementation = "GN.guess")), the estimates of a linear model (for both hyperbolic corrections) and of a cubic model (for the cubic correction with defined minimum- and maximum values (minmax = TRUE)) are computed for initializing the nls (see details below).
The third option is to fit nls with the Levenberg-Marquardt algorithm (using the implementation from the minpack.lm R package). In this case, the start estimates of the nls model are also guessed using either a linear or a cubic model (as previously described).

GN.paper

Algorithm: Gauss-Newton

Parameterizing nls2::nls2() with starting values:

• hyperbolic equation: a = b = d = c(-1000, 1000)
• hyperbolic equation (minmax): b = c(-1000, 1000)
• cubic equation (minmax): a, b = c(-1000, 1000)

options(rBiasCorrection.nls_implementation = "GN.paper")

GN.guess

Algorithm: Gauss-Newton

Parameterizing nls2::nls2() with starting values:

• hyperbolic equation: fitting a linear regression and taking the intercept and the beta as starting values and defaulting d to 1000
• hyperbolic equation (minmax): fitting a linear regression and taking the beta as starting value
• cubic equation (minmax): fitting a cubic regression and taking the betas for the cubic and the quadratic term as starting values

options(rBiasCorrection.nls_implementation = "GN.guess")

LM

Algorithm: Levenberg-Marquardt

Parameterizing minpack.lm::nlsLM() with starting values: same as guessing starting values for option GN.guess

options(rBiasCorrection.nls_implementation = "LM")

BiasCorrector

The GUI BiasCorrector provides the functionality implemented in rBiasCorrection in a web application. For further information please visit https://github.com/kapsner/BiasCorrector.

FAQ

For further information, please refer to the frequently asked questions.

Citation

L.A. Kapsner, M.G. Zavgorodnij, S.P. Majorova, A. Hotz‐Wagenblatt, O.V. Kolychev, I.N. Lebedev, J.D. Hoheisel, A. Hartmann, A. Bauer, S. Mate, H. Prokosch, F. Haller, and E.A. Moskalev, BiasCorrector: fast and accurate correction of all types of experimental biases in quantitative DNA methylation data derived by different technologies, Int. J. Cancer. (2021) ijc.33681. doi:10.1002/ijc.33681.

@article{kapsner2021,
  title = {{{BiasCorrector}}: Fast and Accurate Correction of All Types of Experimental Biases in Quantitative {{DNA}} Methylation Data Derived by Different Technologies},
  author = {Kapsner, Lorenz A. and Zavgorodnij, Mikhail G. and Majorova, Svetlana P. and Hotz-Wagenblatt, Agnes and Kolychev, Oleg V. and Lebedev, Igor N. and Hoheisel, J{\"o}rg D. and Hartmann, Arndt and Bauer, Andrea and Mate, Sebastian and Prokosch, Hans-Ulrich and Haller, Florian and Moskalev, Evgeny A.},
  year = {2021},
  month = may,
  pages = {ijc.33681},
  issn = {0020-7136, 1097-0215},
  doi = {10.1002/ijc.33681},
  journal = {International Journal of Cancer},
  language = {en}
}

More Infos

• Original work by Moskalev et al.: https://doi.org/10.1093/nar/gkr213