ebCoexpressSeries: Functions to run single-study analyses
In EBcoexpress: EBcoexpress for Differential Co-Expression Analysis

Description Usage Arguments Details Value Author(s) References Examples

Core functions that run single-study DC analyses in EBcoexpress

1
2
3

ebCoexpressFullTCAECM(D, conditions, pattern, hpEsts, controlOptions = list())
ebCoexpressOneStep(D, conditions, pattern, hpEsts, controlOptions = list())
ebCoexpressZeroStep(D, conditions, pattern, hpEsts, controlOptions = list())

`D`	The correlation matrix output of makeMyD()
`conditions`	The conditions array
`pattern`	An appropriate output from ebPatterns()
`hpEsts`	The initial hyperparameter estimates from initializeHP() or some other method
`controlOptions`	A list with many options for controlling execution: These two are common to all members of the series: applyTransform: Should Fisher's Z-transformation be applied? Defaults to TRUE verbose: Controls auto-commenting; set to 0 to turn off comments These are used only by the OneStep and FullTCAECM versions: convtol: Convergence tolerance for the EM; default is 5e-04 enforceFloor: Should EC proportion never drop below 0.8? Default is TRUE subsize: If non-NULL, a value less than the 1st dimension of D (p). The EM will use subsize randomly chosen pairs in its computations instead of all p pairs; by default, all pairs are used. We suggest use of this option when the number of pairs is very large m2MaxIter: Upper limit of the number of itertions the optimizer from minqa will use in the M2 step; defaults to 100. For unrestricted iterations, set this to NULL

These three functions represent different flavors of the TCAECM. The FullTCAECM version will run a full TCAECM. The OneStep version will perform a single iteration of the TCAECM and return the results. The ZeroStep version does not perform any EM calculations and instead uses the initial estimates of the hyperparameters to generate posterior probabilities of DC.

We recommend using the OneStep version in most cases.

The output is a list with two members, MODEL and POSTPROBS:

MODEL is a list containing an array MIX and a list HPS. MIX contains estimated mixing proportions for EC/DC classes. HPS contains model specifications:

`G`	The number of mixture components (1, 2 or 3)
`MUS`	Means across mixture components
`TAUS`	Standard deviations across mixture components
`WEIGHTS`	Weights across mixture components (these sum to 1)

The list of lists required by ebCoexpressMeta() is obtained by listing the separate analyses' HPS lists

POSTPROBS is a p-by-L matrix containing posterior probabilities of EC and DC over all L EC/DC classes. The EC posterior probabilities will always be in the first column (which should be fed into crit.fun() if using the soft threshold). Total posterior probabilities of DC for each gene pair are found by summing over the other L-1 columns (or taking 1 minus the first (EC) column)

John A. Dawson <jadawson@wisc.edu>

Dawson JA and Kendziorski C. An empirical Bayesian approach for identifying differential co-expression in high-throughput experiments. (2011) Biometrics. E-publication before print: http://onlinelibrary.wiley.com/doi/10.1111/j.1541-0420.2011.01688.x/abstract

data(fiftyGenes)
tinyCond <- c(rep(1,100),rep(2,25))
tinyPat <- ebPatterns(c("1,1","1,2"))
D <- makeMyD(fiftyGenes, tinyCond, useBWMC=TRUE)
set.seed(3)
initHP <- initializeHP(D, tinyCond)

zout <- ebCoexpressZeroStep(D, tinyCond, tinyPat, initHP)
## Not run: oout <- ebCoexpressOneStep(D, tinyCond, tinyPat, initHP)
## Not run: fout <- ebCoexpressFullTCAECM(D, tinyCond, tinyPat, initHP)

softThresh <- crit.fun(zout$POSTPROB[,1], 0.05)