1175P - Universal circulating tumor DNA quantification using deep learning

Background

The ability to monitor tumor growth dynamics based on circulating tumor DNA (ctDNA) levels in the blood provides a promising non-invasive approach to track disease progression during therapy and clinical trials. Ultra-deep targeted cell-free DNA (cfDNA) sequencing assays are often preferred in the clinic due to their ability to identify actionable mutations. While mutation variant allele frequencies (VAFs) can be used to approximate ctDNA levels and kinetics, not all tumors will have mutations covered by a given targeted sequencing gene panel. We developed Fragle, a multi-stage machine learning model that complements existing targeted sequencing assays by quantifying ctDNA levels directly from the raw cfDNA fragment length density distribution.

Methods

We developed Fragle using low-pass whole genome sequence (lpWGS) and targeted panel sequencing data from 8 cancer types and healthy control cohorts. Using an in silico data augmentation approach, we trained and evaluated Fragle on more than 2500 lpWGS samples across distinct cancer types and healthy cohorts. We evaluated the accuracy and the lower limit of detection (LoD) in independent cohorts and cancer types.

Results

Fragle demonstrated high accuracy and improved lower limit of detection in independent cohorts as compared to existing tumor-naïve ctDNA quantification methods. We tested the method on data from targeted sequencing panels, demonstrating high accuracy across both research and commercial targeted gene panels. We used this feature to study dense longitudinal plasma samples from colorectal cancer patients, identifying strong concordance of ctDNA dynamics and treatment response. To explore the use of Fragle for detection of minimal residual disease (MRD), we analyzed ctDNA levels in a cohort of 162 resected lung cancer patients with plasma profiled using a commercial targeted sequencing panel at the landmark timepoint (∼30 days following surgery). In this cohort, the method significantly improved risk stratification beyond a tumor-naïve targeted panel.

Conclusions

Overall, Fragle is a versatile, fast, and accurate method for ctDNA quantification with potential for broad clinical utility.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

National Medical Research Council (NMRC), Singapore.

Disclosure

All authors have declared no conflicts of interest.