mogsa (version 1.6.4)

moa: Multiple omics data analysis using MFA or STATIS

Description

Analysis multiple omics data using MFA or STATIS. The input multiple tables are in a form that columns are samples and rows are variables/features.

Usage

moa(data, proc.row="center_ssq1", w.data="inertia", w.row=NULL, statis=FALSE)

Arguments

data

A list of data.frame or matrix that contains the input datas, the columns in all datasets should be samples/observations (which need to be matched) and rows should be variables.

proc.row

Preprocessing of rows of datasets, should be one of none - no preprocessing, center - center only, center_ssq1 - center and scale (sum of squred values equals 1), center_ssqN - center and scale (sum of squred values equals the number of columns), center_ssqNm1 - center and scale (sum of squred values equals the number of columns - 1) MFA corresponds to "proc.row=center_ssq1" and 'w.data="lambda1"'

w.data

The weights of each separate dataset, should be one of uniform - no weighting,

lambda1 - weighted by the reverse of the first eigenvalue of each individual dataset

or inertia - weighted by the reverse of the total inertia. See detail.

w.row

If it is not null, it should be a list of positive numerical vectors, the length of which should be the same with the number of rows of each dataset to indicated the weight of rows of datasets.

statis

A logical indicates whether STATIS method should be used. See details.

Value

An object of class moa-class.

Details

Different methods employs different precessing of row and datasets. For multipple factorial analysis (MFA), the rows of each dataset are first centered and scaled, then each dataset is weighted by the reverse of its first eigenvalue (proc.row=center_ssq1, w.data="lambda1"). This algorithm does not have a well defined criterion to be optimized (see reference).

If statis=TRUE, the statis algorithm will be used, that is, each dataset will be further weighted so that datasets closer to the overall structure will receive a higher weight.

References

Herve Abdi, Lynne J. Williams, Domininique Valentin and Mohammed Bennani-Dosse. STATIS and DISTATIS: optimum multitable principal component analysis and three way metric multidimensional scaling. WIREs Comput Stat 2012. Volume 4, Issue 2, pages 124-167 Herve Abdi, Lynne J. Williams, Domininique Valentin. Multiple factor analysis: principal component analysis for multitable and multiblock data sets. WIREs Comput Stat 2013

See Also

sup.moa, mogsa. More about plot see moa-class.

Examples

  # library(mogsa)
  # loading data
  data(NCI60_4arrays)
  # run analysis
  ana <- moa(NCI60_4arrays, proc.row = "center_ssq1", w.data = "inertia", statis = TRUE)
  # plot
  # plot eigen value
  plot(ana, value = "eig", type = 2)
  # plot the normalized (percentage) eigen value
  plot(ana, value = "tau", type = 2)
  # ploting the observations
  colcode <- as.factor(sapply(strsplit(colnames(NCI60_4arrays$agilent), split="\\."), "[", 1))
  plot(ana, type = 1, value = "obs", col=colcode)
  plot(ana, type = 2, value = "obs", col=colcode, data.pch=1:4)
  # plot variables/features in each data sets
  plot(ana, value = "var", layout=matrix(1:4, 2, 2))
  # plot the RV coefficients for the data sets
  plot(ana, value = "RV")