This technology is a framework that uses machine learning to identify oncogenic somatic and germline mutations in tumor samples for personalized cancer therapy.
Cancer results from specific genetic mutations, and identification of cancer-causing mutations underlies current genome-based precision cancer treatment. DNA variants can be either somatic (tumor-/tissue-specific), or germline (hereditary), and determination of the origin and location of these variants is important in developing targeted cancer therapeutics. Despite the value in sequencing DNA to differentiate germline and somatic variants, the high cost of sequencing has led to tumor-only sequencing in both research and clinical settings. Currently, there are no techniques available to integrate information from both the individual patient sequences and the total patient cohort.
This technology describes a framework to learn features of recurrent somatic mutations to predict somatic variants from tumor-only samples and identify somatic-like germline variants for integrated analysis of tumor-normal DNA. Tumor and germline DNA from a small number of patients are used to classify variants in tumor-only samples, enabling analysis to be done on largely tumor-only samples. This framework, called Tumor-Only Boosting Identification (TOBI), uses a gradient boosting algorithm to learn from confirmed somatic variants and classify variants with somatic characteristics. Results have demonstrated that TOBI has correctly recognized 90% of somatic variants. Additionally, TOBI identified variants in BRCA2 and other Fanconi anemia genes in 11% of bladder cancers, suggesting a genetic predisposition and potential for treatment with poly ADP-ribose polymerase (PARP) inhibitors. This technology has the potential to help identify which DNA variants are responsible for the development of various cancers to enable the identification of more targeted treatments.
This technology has been validated on tumor samples from 1769 patients from 7 cancer types.
Patent Pending (WO/2018/064547)
IR CU17071
Licensing Contact: Ron Katz