| Title: | Measurement Error model estimate for correlation coefficient |
|---|---|
| Description: | Two-stage measurement error model for correlation estimation with smaller bias than the usual sample correlation |
| Authors: | Beiying Ding |
| Maintainer: | Beiying Ding <[email protected]> |
| License: | LGPL |
| Version: | 1.77.0 |
| Built: | 2024-09-29 06:04:18 UTC |
| Source: | https://github.com/bioc/MeasurementError.cor |
Given a matrix ( p x n) for observed values of p variables and a corresponding matrix for their standard errors, the all pairwise measurement error estimates for true correlations are returned
cor.me.matrix(exp, se)cor.me.matrix(exp, se)
exp |
observed value marix |
se |
standard error matrix |
The final estimates for true correlation (i.e. cor.true) from the measurement error model
The function involves using quasi-newton for linear optimization, "BFGS" is the only implemented method now.
Refer to cor.me.vector for more details.
Beiying Ding
Ding, B.Y. and Gentleman, R.(2003) Measurement error model for correlation coeffcient estimation and its application in microarray analsysis
cor.me.vector
exp <- matrix(abs(rnorm(200,1000,20)),ncol=10) se <- matrix(abs(rnorm(200,50,5)),ncol=10) cor.me.matrix(exp,se)exp <- matrix(abs(rnorm(200,1000,20)),ncol=10) se <- matrix(abs(rnorm(200,50,5)),ncol=10) cor.me.matrix(exp,se)
Given the observed value of two variables and their respective standard error, the measurement error estimate for their correlation coefficient is returned
cor.me.vector(exp1, se1, exp2, se2)cor.me.vector(exp1, se1, exp2, se2)
exp1 |
observed value for vector 1 |
se1 |
estimated standard error for vector 1 |
exp2 |
observed value for vector 2 |
se2 |
estimated standard error for vector 2 |
estimate |
Vecotr containing the estimates from the measurement error model, i.e. |
count |
numer of function and gradient evaluation |
convergence |
0 if converged. See optim() for details |
Most applicable for microarray expression data where standard errors are readily estimated by most low level analysis softwares. Hence variables can be thought of as genes. One also need to differentiate between cor.me and cor.true: the first one being the correlation between the measurement error distributions of the two genes whereas the second one is the quantity of interest, i.e true correlation between the two gene expression profiles.\
The function involves using quasi-newton for linear optimization, "BFGS" is the only implemented method now.
Beiying Ding
Ding, B.Y. and Gentleman, R. (2003) Measurement Error Model for correlation coefficient estimation and its application in microarray analysis
cor.me.matrix
exp <- matrix(abs(rnorm(200,1000,20)),ncol=10) se <- matrix(abs(rnorm(200,50,5)),ncol=10) cor.me.vector(exp[1,],se[1,],exp[2,],se[2,])exp <- matrix(abs(rnorm(200,1000,20)),ncol=10) se <- matrix(abs(rnorm(200,50,5)),ncol=10) cor.me.vector(exp[1,],se[1,],exp[2,],se[2,])