Abstract 4294: Bridging the gap between targeted NGS and FISH gene-level CNV detection capabilities in hematologic malignancies

Abstract

Abstract Background: Copy Number Variations (CNVs) are prominent features of cancer cells. From a clinical standpoint, their accurate detection at a low cost is a priority. With regular increases in the number of markers to be tested, the cost effectiveness and practicality of gold standard techniques like Fluorescence In Situ Hybridization (FISH) are slowly decreasing. Cost-efficient Next Generation Sequencing (NGS) targeted gene panels can be scaled up but accurately detecting CNVs from the resulting data remains challenging. We demonstrate large amounts of data and machine learning can help bridge the gap between the two techniques. Methods: We collected the sequencing data of 6,277 patients tested using a custom amplicon based NGS assay designed to detect somatic alterations in 297 hematological cancer relevant genes such that at least one concurrent FISH test was also performed. FISH results were used to infer the gain, loss, or normality information for each corresponding gene. The annotated genes were then used to curate a training set by extracting 20 features per gene from the alignment results. A 3-class random forest classifier was trained using this dataset. The selected model was evaluated on a distinct set of 2,738 patients. Results: Evaluation results are provided in Table 1 for 8 genes for which the FISH probe used to infer the gain, loss, or normality information directly spanned the gene region. The predicted CNVs are almost a perfect match with FISH for 6 of these genes with a limit of detection at 20% abnormal cells. In most cases, the model reduces discordant calls by over 50% compared to using existing CNV detection software only. Conclusion: We show the CNV detection capabilities of a targeted NGS assay can closely match the gold-standard FISH technique by analytically correcting the biases introduced by the targeting procedures. The model presented here is used to detect CNVs in ALL patients after a successful formal validation in our laboratory. Table 1. Evaluation results. Gene FISH Positive Cases (Gain/Loss) FISH Negative Cases (Normal) All FISH Cases Total Concordant Sensitivity Total Concordant Specificity Total Concordant Accuracy ATM 72 72 100.00% 641 629 98.13% 713 701 98.32% CBFB 23 22 95.65% 517 506 97.87% 540 528 97.78% EGR1 171 169 98.83% 1,541 1,538 99.81% 1,712 1,707 99.71% KMT2A 27 25 92.59% 474 472 99.58% 501 497 99.20% MET 113 113 100.00% 1,593 1,582 99.31% 1,706 1,695 99.36% NF1 15 15 100.00% 908 908 100.00% 923 923 100.00% TERT 10 10 100.00% 1,723 1,709 99.19% 1,733 1,719 99.19% TP53 76 73 96.05% 1,567 1,556 99.30% 1,643 1,629 99.15% Citation Format: Christophe N. Magnan, Hyunjun Nam, Shashikant Kulkarni, Segun C. Jung. Bridging the gap between targeted NGS and FISH gene-level CNV detection capabilities in hematologic malignancies. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4294.