Abstract 5015: Precision profile simulation study for a next generation sequencing bTMB assay

• Published in: Cancer research (Chicago, Ill.) Ill.), 2022-06, Vol.82 (12_Supplement), p.5015-5015
• Main Authors: Doubleday, Kevin, Gaile, Daniel, Vijaya-Satya, Ravi, Liu, Xianxian, D'Auria, Kevin, Shukla, Soni, Chuang, Han-Yu, Quinn, Katie, Chudova, Darya
• Format: Article
• Language: English
• ISSN: 1538-7445; 1538-7445
• DOI: 10.1158/1538-7445.AM2022-5015

Background: Precision profile simulations (PPS) can be used to assess variability of biomarker profiles and provide valuable insight into assay performance, especially when reliable precision estimates can not be obtained empirically due to scarcity of representative samples or insufficient materials per sample. A PPS was conducted for the GuardantOMNI assay to characterize the expected variability in blood tumor mutational burden (bTMB) score across a representative range of expected bTMB scores in clinical samples. The simulations were aligned to, but not completely prescribed by, the PPS guidance provided in Guidance for Industry and and Food and Drug Administration Staff Class II Special Controls Guidance Document: Ovarian Adnexal Mass Assessment Score Test System. A sample’s bTMB score is a real valued quantity (e.g., bTMB = 21.04 mut/Mb) that is derived by multiplying the number of qualified mutations observed within a targeted panel by a scaling factor. Variability in observed bTMB scores for a given blood sample is governed primarily by sample coverage, tumor shedding level, and the assay somatic variant detection probabilities (a function of underlying variant allele frequencies, VAFs). Methods: The relationship between site-specific total molecule counts and coverage was modeled utilizing a composite dataset consisting of both clinical and contrived samples. Sample coverage was modeled using variance component estimates from Precision Study data (18 cancer samples each with 6 to 18 replicates). The reference, single-strand mutant, and double-strand mutant molecule counts for a somatic variant site detected in at least one sample replicate were modeled utilizing a bias corrected Dirichlet Multinomial model. The variants with the simulated VAF and coverage levels were processed with the GuardantOMNI germline/somatic classifier to account for the uncertainty in germline/somatic classification at lower coverage values. Results: Precision profiles consisting of simulation derived %CV estimates for 18 clinical samples with a representative set of mean bTMB scores were generated. The PPS bTMB score distributions were consistent with the bTMB scores observed in the Precision Study, supported by visualization and confidence intervals at level 0.05 margins of equivalence for the empirical mean bTMB and standard deviation estimates. The sample specific %CV estimates were observed, in most instances, to decrease with increasing input levels for matched targe