Yaron Einhorn leads the bioinformatics and research in Genoox since 2016.
Yaron Einhorn, Moshe Einhorn, Yaron Assa, Odem Shani
Genoox, Tel-Aviv, Israel
Introduction: The ClinGen, CGC and VICC recently published new recommendations for the classification of pathogenicity of somatic variants in cancer, in order to create a set of standards when classifying the oncogenicity of a somatic variant. While the detailed evaluation criteria will increase classification consistency between different labs, the process is time-consuming and holds some computational challenges. Here, we present a novel AI-based oncogenicity classification engine for somatic variants based on these new recommendations and integrated it into Franklin (franklin.genoox.com), an open-access platform for variant interpretation.
Methods: An AI oncogenic classification engine was developed and integrated into Franklin for implementing the new recommendations and removing their computational and technical challenges. We evaluated the classifications against the dataset which was classified by experts during the process of creating these recommendations, which included 94 variants in 10 cancer-related genes.
Results: Comparing the results shows high concordance without any strong conflict between the experts and Franklin. 100% (34/34) of the Oncogenic/Likely-Oncogenic variants were classified as Oncogenic/Likely-Oncogenic, 100% (7/7) of the B/LB variants were classified as B/LB/VUS-Leaning-Benign (VUS-B). 74% (23/31) of the VUS variants were classified as VUS, VUS-Leaning-Oncogenetic, or VUS-B. Of the 8 VUS conflicts, 2 were classified as B/LB, and 6 were classified as Oncogenic/Likely-Oncogenic, suggesting that there might be additional evidence to be considered.
Conclusions: Herein we demonstrated how Franklin can aid in following the new oncogenic classification recommendations, a task that is challenging, and time-consuming. To the best of our knowledge, this is the only solution that currently exists today.