gedAlgorithm.deconf. CellMix 1.6

Description

The ged method ‘deconf’ uses an alternate least-squares algorithm to estimate both cell proportions and cell-specific signatures from global expression data, as proposed by Repsilber et al. (2010).

Usage

gedAlgorithm.deconf(target, x, maxIter = 1000L, error.threshold = 0, fit = c("fast", 
  "original"), ...)

Arguments

fit: fitting method: 'fast' uses fcnnls, while 'original' uses the original implementation from the deconf package (see Details).
...: other arguments passed to nmf. Note that argument data is not allowed when x is a MarkerList object.
target: target mixed expression matrix.
x: factorisation rank, i.e. the number of cell types to extract, or a complete initial NMF model.
maxIter: maximum number of iterations
error.threshold: min error to continue iterations, if error falls below this threshold, the iterative algorithm stops

Details

This method fits an NMF model to the data in a completely unsupervised manner. If marker genes are provided, they are used a posteriori to assign each estimated component, i.e. each cell-specific signature, to the cell-type with the greatest proportions of consistent markers.

Fast built-in implementation

The method ‘deconf’ is implemented as an NMF algorithm, which is registered under the same names in the NMF package's algorithm registry.

It uses an improved implementation, based on the fast combinatorial nonnegative least-squares algorithm from Van Benthem et al. (2004), Kim et al. (2007), as provided by the function fcnnls in the NMF package. This enables to achieve great performance speed-up, being really -- way -- much faster than the original implementation.

Original implementation

The CellMix also includes a way to run the original version from the deconf package, using argument fit = 'original'.

This version requires the deconf package, which was released as supplementary data only to support the paper from Repsilber et al. (2010), i.e. it is not available from CRAN or Bioconductor. However, we made it available from the CellMix CRAN-like support repository:

http://web.cbio.uct.ac.za/~renaud/CRAN

The easiest way to install it is to run:

install.extras('CellMix', 'deconf')

References

Repsilber D, Kern S, Telaar A, Walzl G, Black GF, Selbig J, Parida SK, Kaufmann SHE and Jacobsen M (2010). "Biomarker discovery in heterogeneous tissue samples -taking the in-silico deconfounding approach." _BMC bioinformatics_, *11*, pp. 27. ISSN 1471-2105, , .

Van Benthem M and Keenan MR (2004). "Fast algorithm for the solution of large-scale non-negativity-constrained least squares problems." _Journal of Chemometrics_, *18*(10), pp. 441-450. ISSN 0886-9383, , .

Kim H and Park H (2007). "Sparse non-negative matrix factorizations via alternating non-negativity-constrained least squares for microarray data analysis." _Bioinformatics (Oxford, England)_, *23*(12), pp. 1495-502. ISSN 1460-2059, , .