MBASIC.binary: Bayesian clustering model for binary state matrix with prior...

In chandlerzuo/mbasic: Matrix Based Analysis for State-space Inference and Clustering.

Description

Bayesian clustering model for binary state matrix with prior estimated background means. This function is deprecated.

Usage

MBASIC.binary(Y, Mu0, fac, J = NULL, zeta = 0.2, maxitr = 100,
  burnin = 20, outfile = NULL, out = NULL, init.mod = NULL,
  struct = NULL, family = "lognormal", tol = 1e-04, nsig = 1,
  min.count = 5)

Arguments

Y	An N by I matrix containing the data from N experiments across I observation units (loci).
Mu0	An N by I matrix for the prior estimated mean for the background state, for N experiments across the I observation units (loci).
fac	A vector of length N denoting the experimental condition for each replicate.
J	The number of clusters to be identified.
zeta	The initial value of the proportion of unclustered units. Default: 0.2.
maxitr	The maximum number of iterations in the E-M algorithm. Default: 100.
burnin	An integer value for the number of iterations in initialization. Default: 20.
outfile	The file directory for writing the intermediate results every 10 E-M iterations. This can be useful when the running time until final convergence is long. Default: NULL (no intermediate result is saved).
out	The file directory for writing fitting information in each E-M iteration. Default: NULL (no information is outputted).
init.mod	A 'MBASICFit' class object taken as the initial model to initialize parameters. This object can be an intermediate result from a not yet converged E-M algorithm, or a fitted model with smaller number of clusters. The user must be cautious in providing this initial model, since it must be fitted using the same data as the argument of this function.
struct	A matrix indicating the levels of the signal matrix.
family	The distribution of family to be used. Either "lognormal" or "negbin". See details for more information.
tol	Tolerance for error in checking the E-M algorithm's convergence. Default: 1e-04.
nsig	The number of mixture components for the distribution of the signal state.

Details

Function MBASIC.binary currently supports two different distributional families: log-normal and negative binomial. This should be specified by the 'family' argument.
For the log-normal distributions, log(Y+1) is modeled as normal distributions. For experiment n, if locus i is unenriched, distribution for log(Y[n,i]+1) is N(e[n] * Mu0[n,i], sigma0[n]). If locus i is enriched and the enrichment state is s, the distribution of log(Y[n,i]+1) is N(mu1[n,s], sigma1[n,s]).
For the negative binomial distributions, the meanings of mu1, sigma1, sigma0 changes. For experiment n, if locus i is unenriched, distribution of Y[n,i] is NB(Mu0[n,i] * e[n], sigma0[n]). Otherwise, if locus i is enriched and the enrichment component is s, the distribution of Y[n]-1 is NB(mu1[n,s], sigma[n,s]). In this package, NB(mu, a) denotes the negative-binomial distribution with mean mu and size a (i.e. the variance is mu*(1+mu/a)).

Value

A 'MBASICFit' class object.

Author(s)

Chandler Zuo [e-mail]

Examples

## Not run: 
## Simulate a data
dat <- MBASIC.sim.binary(I = 100, fac = rep(1:5, each = 2), J = 3, f = 5, family = "lognormal")
## Fit the model
fit <- MBASIC.binary(t(dat$Y), Mu0 = t(dat$Mu0), fac = rep(1:5, each = 2), J=3, struct = NULL, family="lognormal")

## End(Not run)

chandlerzuo/mbasic documentation built on May 13, 2019, 3:24 p.m.