1246P - Detection of alternative lengthening of telomeres (ALT) across cancer types based on tumor-normal multigene panel sequencing

Background

Alternative Lengthening of Telomeres (ALT) is a telomerase-independent mechanism for the maintenance of telomeres, which is characterized by the absence of telomerase activity and often co-occurring loss of function somatic mutations in ATRX or DAXX. There is burgeoning evidence to suggest that ALT is associated with genetic vulnerabilities that may be exploited through synthetically lethal approaches. Here, we sought to develop a classifier to detect ALT based on multi-gene targeted sequencing data.

Methods

We developed an attention-based neural network model to detect ALT based on a clinical tumor-normal panel sequencing data (MSK-IMPACT). For training, cases with deleterious ATRX or DAXX mutations were considered ALT (n=274), whereas cases harboring TERT promoter hotspot mutations or gene amplification were regarded as non-ALT (n=965). The model was trained using a 10-fold cross-validation and 997 genetic and clinical features, including cancer type, telomeric content and fusion rates, copy number variation and mutational profiles, followed by hyperparameters optimization. The model was independently validated in 619 cases across 36 cancer types with an ALT:non-ALT ratio of 1:3.

Results

In the validation cohort, the MLP classifier detected ALT cancers with an accuracy of 80% (99% CI 75.9-84.1%), a sensitivity of 69% (99% CI 64.2-74.8%), specificity of 88% (99% CI 84.6-91.4%), and a F1 score of 75% (99% CI 70.5-79.5%). In the context of samples correctly classified (n=495), ALT was significantly more frequently detected in GI neuroendocrine tumors, uterine sarcomas, as well as adrenocortical, prostate and small cell lung cancer.

Conclusions

Deep learning can detect ALT cancers from targeted sequencing data across various cancer types. This new method enables the dissection of genomic and phenotypic characteristics of ALT cancers in large retrospective cohorts with available targeted sequencing data, while providing a potentially high-throughput, accessible means to enroll patients in prospective studies targeting the ALT phenotype.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Memorial Sloan Kettering Cancer Center.

Funding

NIH K12.

Disclosure

B.H. Diplas: Non-Financial Interests, Personal and Institutional, Research Funding: NIH K12, MSK FGI. B. Weigelt: Financial Interests, Personal, Funding: Repare Therapeutics. E. Rosen: Financial Interests, Personal and Institutional, Research Funding: Bayer. A. Sfeir: Financial Interests, Personal, Advisory Board, Shareholder, Co-Founder, Consultant: Repare Therapeutics. J.S. Reis-Filho: Financial Interests, Personal, Other, Consultant: Goldman Sachs, Eli Lilly, Saga Diagnostics; Financial Interests, Personal, Other, Member of the Scientific Advisory Board and Consultant: Repare Therapeutics, Paige.AI; Financial Interests, Personal, Advisory Board: Personalis, Roche Tissue Diagnostics; Financial Interests, Personal, Advisory Board, Member of the Scientific Advisory Board: Bain Capital; Financial Interests, Personal, Advisory Board, Ad hoc member of the Pathology Scientific Advisory Board: Daiichi Sankyo, Merck; Financial Interests, Personal, Advisory Board, Ad hoc member of the Oncology Scientific Advisory Board: AstraZeneca; Financial Interests, Personal, Advisory Board, Member of the SAB: MultiplexDX; Financial Interests, Personal, Member of Board of Directors: Odyssey Bio, Grupo Oncoclinicas; Financial Interests, Personal, Stocks/Shares: Repare Therapeutics; Financial Interests, Personal, Other, Stock options: Paige.AI. All other authors have declared no conflicts of interest.