Abstract 6301: ARLVPI: A novel machine-learning-informed platform for assessing somatic variant classification and pathogenicity

Abstract Cancer comprises several heterogeneous multifaceted diseases characterized by dysfunctional cellular differentiation and excessive proliferation. With its prevalence expected to rise, particularly among middle-aged adults, the burden on pathologists to identify and characterize pathogenic variants increases. Current clinical practices rely on molecular profiling, pathologist interpretation, and supplemental information from disease databases to classify these variants. In this context, variants of unknown significance (VUS) pose a significant clinical challenge due to their uncertain biological and clinical effects that make them inhibitive towards diagnoses and treatment decisions. Representing 40% of Next-Generation Sequencing (NGS) signals, VUS indicates a significant unmet clinical need demanding urgent solutions. To meet this unmet clinical challenge, we utilized 20, 756 variants from 415 cancer patient samples to develop the Advanced Risk Locating Variant Pathogenicity Index (ArlVPI). ArlVPI employs a mechanistic model based on information from a random forest classifier and dimensional reduction analyses to identify and flag VUS variants likely to require further investigation from pathologists. In downstream analysis, ArlVPI exhibited a strong correlation with variant pathogenicity and correctly categorized variants as benign or pathogenic with <1% error. With respect to VUS, ArlVPI flagged 16% as requiring potential follow-up, corroborating metrics observed in literature. Furthermore, ArlVPI demonstrated >93% concordance with clinical pathologists in the identification of VUS which were reclassified as pathogenic in an array of solid tumor, hematologic, and myeloproliferative malignancies. Our study illustrates how ArlVPI can be integrated into existing NGS analysis pipelines to support molecular professionals, offering an efficient resource that enhances variant classification and supports targeted cancer treatment strategies. Future work focuses on developing a streamlined application that interfaces with existing clinical diagnostic systems. Citation Format: Arlen Larry Gyden, Nicole Ambroise, Tarsus Lam, Zehra Ordulu, Petr Starostik, Kimberly Newsom, Mohammed Gbadamosi. ARLVPI: A novel machine-learning-informed platform for assessing somatic variant classification and pathogenicity [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 6301.