Inferring the location of tumor suppressor genes by modeling frequency of allelic loss

Allelic loss is often part of a multistep process leading to tumorigenesis. Analysis of genomic markers highlights regions of elevated allelic loss, which in turn suggests a nearby tumor suppressor. Furthermore, pooling published analyses to combine evidence can increase the power to detect a tumor suppressor gene. If the pattern of loss for each tumor, or allelotype, is known, a stochastic model proposed by Newton et al. (1998, Statistics in Medicine 17, 1425-1445) can be used to analyze the correlated binary data. Many studies report only incomplete allelotypes, augmented with frequencies of allelic loss (FAL) at each marker, in which the number of informative tumors showing allelic loss is provided along with the number of informative tumors. We describe an extension of the allelotype model to handle FAL data, using a hidden Markov model or a normal approximation to compute the likelihood. The FAL model is illustrated using data from a study of colorectal cancer.