Amber, a junior studying Computational Biology at the University of South Carolina, benchmarked Tonly2, a somatic-germline mutation classifier using tumor-only samples.
Mutations in the genome can play an important role in cancer development. Spending my summer at the Broad was an incredible experience. Before coming to the Broad, I could not have imagined how much growth I would have in nine weeks or how intellectually stimulated or challenged I would be. I was exposed to the bleeding edge of computational cancer genomics and immunology. Furthermore, I had the best mentorship experience I could have asked for. There are two types of mutations in cancer: acquired somatic mutations and inherited germline mutations. Since somatic and germline mutations arise from different mechanisms, it is imperative to differentiate between them. Furthermore, these classifications are important for downstream analyses, such as identifying acquired cancer driver mutations or inherited cancer risk alleles. This identification is typically done by comparing the mutations present in the tumor to those in normal cells. However, normal samples are sometimes unavailable or have high contamination of tumor cells. Without a normal sample, identifying germline mutations is no longer a trivial task. Moreover, germline mutations outnumber somatic mutations in most tumor samples, making it challenging to identify germline mutations without misclassifying true somatic mutations.
Tonly2 is a computational method developed to classify mutations as somatic or germline from next-generation sequencing (NGS) data of tumor-only samples while controlling for sensitivity (percentage of true somatic mutations classified as somatic). Tonly2 compares the observed variant allele fraction (proportion of sequence reads supporting a mutation, VAF) to the expected VAF of a germline or somatic mutation, which is calculated based on purity (proportion of tumor cells in a tumor sample), local copy number (amplifications or deletions of segments in the genome), ploidy (number of chromosome sets), and cancer cell fraction (proportion of tumor cells containing variant). Tonly2 then uses a log odds ratio to determine whether the data best fits a somatic or germline model. Tonly2 was evaluated on 84 whole-genome lung adenocarcinoma samples with matched normal samples, and its performance was compared to SGZ, a similar tumor-only method. Tonly2 classified more germline mutations correctly compared to SGZ (Tonly2 median 79.77%; SGZ median 67.58%) while maintaining a median sensitivity of 96.07%.
Project: Tonly2: A Somatic-Germline Mutation Classifier Using Tumor-Only Samples
Mentors: Claudia Chu & Kristy Schlueter-Kuck
PI: Getz Lab, Cancer Program