Inferring the location and effect of tumor suppressor genes by instability-selection modeling of allelic-loss data

Biometrics , 56, 1088-1097, 2000.

(Originally issued May 7, 1999, as Department of Biostatistics and Medical Informatics Technical Report #135.)

Abstract:

Cancerous tumor growth creates cells with abnormal DNA. Allelic-loss experiments identify genomic deletions in cancer cells, but sources of variation and intrinsic dependencies complicate inference about the location and effect of suppressor genes; such genes are the target of these experiments and are thought to be involved in tumor development. We investigate properties of an instability-selection model of allelic-loss data, including likelihood-based parameter estimation and hypothesis testing. By considering a special complete-data case, we derive an approximate calibration method for hypothesis tests of sporadic deletion. Parametric bootstrap and Bayesian computations are also developed. Data from three allelic-loss studies are reanalyzed to illustrate the methods.
Key words: Allelic imbalance; Cancer gene mapping; Chromosomal deletions; Correlated binary data; LOD score; Loss of heterozygosity.

This manuscript is a follow up to

M.A. Newton , M.N. Gould, C. A. Reznikoff, and J.D. Haag (1998). On the statistical analysis of allelic-loss data. Statistics in Medicine 17 , 1425-1445.

Data Sets Analyzed (in R/Splus dput/dget format):
Allelic loss data is in the matrix data (LOH=1,MOH=0,no data=-1); marker positions relative to (0,1) are in the vector pos .

Juvenile polyposis coli fjp.data
Thyroid cancer thy.data
Prostate cancer pro.data

R/Splus code

Maximum likelihood estimation

Source for data input: Download one of the data files listed above.

Source code for likelihood evaluation : mloglik.null.s (under the null ), mloglik.alt.s (under the alternative).

Source code for maximum likelihood estimation: mle.null.s (under the null), mle.alt.s (under the alternative).

Under the null hypothesis, log-likelihood function was maximized over lambda, and delta. Under the alternative hypothesis, it was maximized over lambda, delta, omega, and locus under the constraint that omega >= delta. For maximization nlminb (S-plus function) was used.

Source code for simple estimators : simple.est.null.s (under the null), simple.est.alt.s (under the alternative).

See comments in the source files to see how simple estimators were computed.

Source code for plotting profile lod curves: lod.plot.s

Brief instructions: In order to use nlminb (S-plus function), code for parameter estimation should be run on S-plus. The remaining code can be run on either S-plus or R. To run code as a batch job on S-plus, type: Splus BATCH <source.s> <output.out> &.

Bayesian analysis via MCMC

Source for data input: Download one of the data files listed above.
Source code for the missing data proposal, s.q
Main source file s.mcmc
Brief instructions: The R/Splus code provided here was used to implement the Bayesian computations described in the above technical report TR135. To use, download a data file from the above list, the file s.q which performs one of the update steps, and the main file, s.mcmc . Typical use will involve a few tests in which the mcmc object in s.mcmc is set to force a short run. A production run will use much longer settings and may take hours on a fast workstation. I usually run this batch, with a command: R --no-save -v 20 < s.mcmc > s.out & Results of the calculation are stored in the file results .

Bugs reported and fixed