Abstract 3965: An algorithm to detect somatic mutations in cancer specimens using mass spectrometric genotyping

• Published in: Cancer research (Chicago, Ill.) Ill.), 2012-04, Vol.72 (8_Supplement), p.3965-3965
• Main Authors: Hatton, Charles, Niu, Lili, Davis, Matthew, Mondal, Chandrani, Luongo, Lauren, Go, Christina, Jones, Robert T., Roden, Christine, Palescandolo, Emanuele, Hummelen, Paul Van, Garraway, Levi A., Meyerson, Matthew, Hahn, William, MacConaill, Laura E.
• Format: Article
• Language: English
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2012-3965

Although advances in technology and the biological understanding of cancer pathogenesis have brought the goal of personalized cancer medicine closer, important challenges remain in several areas. One such challenge is the reliable and sensitive detection of somatic mutations in clinical tumor specimens. Somatic mutation detection in clinical tumor specimens is confounded by biological variables such as the percentage of tumor cells in a specimen, stromal contamination, sequence quality (particularly in DNA derived from formalin-fixed clinical samples), tumor heterogeneity and ploidy considerations. These and other experimental factors can lead to allele signal ratios very different from the 1:0, 1:1 or 0:1 ratio expected for a classically-defined SNP. Germline genotyping using the Sequenom MALDI-TOF mass spectrometry platform has been used to reliably detect single-nucleotide polymorphisms (SNPs) in a wide variety of samples. The automated genotype calling algorithm included with the Sequenom platform falters, however, when the ratio of allele peak signals for a given assay in a specific sample do not approximate the native algorithm's expectations of frequency, confounding accurate genotype calling by the system. We report here a cancer-specific algorithm for sensitive genotype calling of somatic mutations, based on an independent clustering of raw allele peak data for each assay run, and the identification and stratification of outliers from an empirical background (‘wildtype’) cluster of genotypes. Outliers from this cluster are then evaluated using a number of criteria to determine the quality and performance of the overall assay run, as well as the individual call; these factors include the relative signal-to-noise ratio (SNR) of each allele peak, the ratio of variant allele peak height to peak area, the percentage of signal lost in unextended probe (UEP) and relative distance of a putative mutation call from the wildtype cluster In a scatter plot of peak area values. Based on these criteria, a confidence value is assigned to each candidate outlier, allowing for stratification and prioritization of these candidates for validation. We applied the algorithm to samples run against the our OncoMap genotyping panel, comprised of 460 assays interrogating mutations in 33 cancer-associated genes. Candidate mutations produced by this algorithm correlate strongly with those called by users manually, and use of the algorithm increases the sensitivity and reliabi