A Hidden Markov Model based approach to estimation of allelic imbalances in tumor cells

AACR Annual Meeting-- Apr 14-18, 2007; Los Angeles, CA 4410 Researchers have validated correlations between carcinogenesis and genetic alterations in tumor cells. These alterations involve allelic imbalances, manifested in the form of chromosomal copy number(CN) changes - amplifications, deletions, aneuploidy, loss of heterozygosity (LOH), micro-satellite instability among others. These events often indicate the activation of oncogenes and/or inactivation of tumor suppressor genes (anti-oncogenes). Identification of the loci implicated in these aberrations can generate anchor points which facilitate oncogenomics research. Interpretation of these aberrations, however, is complicated by the following, among others: (a) variations in the form of copy number polymorphisms (CNP) can also occur in normal individuals; (b)CN aberrations might not be accompanied by LOH and vice-versa; (c) preferential aberrations in one allele over the other. Motivated by the above rationale, we have developed a Hidden Markov Model based framework to quantify, total copy number (TCN), allele-specific copy-number (ASCN) changes and LOH. The Affymetrix, Genechip Human Mapping 500K array platform which provides consistently high coverage for Nsp and Sty restriction enzymes, with a median inter-SNP distance of 2.5kbases has been used for this analysis . The framework exists to perform a matched tumor-normal tissue versus un-matched tissue analysis - this allows the de-convolution of copynumber polymorphic events from those of tumor related events. We have performed CN and LOH analyses comparing samples of primary ductal carcinoma with and without metastasis of lymph nodes and this data will be presented. This analysis where the normal and tumor samples are derived from the same patient is contrasted with pools of phenotypically normal controls derived from ethnicity matched as well as unmatched HapMap samples. The purpose behind this analytical strategy is threefold. 1) It allows the identification of sample specific CNPs. 2) It establishes the sensitivity and specificity of the outcome for cases where matched normal tissue might not be available. This analysis design also enables the determination of ethnicity biases introduced via the reference distribution. 3) It aggregates the preferential amplification/deletion of one allele over another as a way to specifically catalog and correlate the allelic imbalances with TCN and LOH estimates in matched tumor-normal samples. Validation of the underlying algorithms, using a titration matrix of 1-5 copies of chromosome X results in a R2~0.951; for deletions, median sensitivity is 0.997at 1% FPR. Correlations with aCGH(SMRT) based detection of CN in the EFGR pathway(SK-BR-3 cell line) results in a R2~ 0.88.