P1379 - Accurate Detection of Esophageal Squamous Cell Carcinoma (ESCC) Using Machine Learning with Methylated DNA Biomarkers

Ke Ma, MD¹, Hua-Ling Tsai, MSc¹, Corey Nolet, PhD², Dennis Gong, ³, Samson Okello, MD⁴, Boniface Lumori, MD⁴, Simran Jit, MBBS¹, Andrew Kalra, BS¹, Ludmila Danilova, PhD¹, Yulan Cheng, MD¹, Stephen J. Meltzer, MD^1
1Johns Hopkins University School of Medicine, Baltimore, MD; ²Capsulomics, Inc., Baltimore, MD; ³Johns Hopkins University, Baltimore, MD; ⁴Mbarara University of Science and Technology, Mbarara, Mbarara, Uganda

Introduction: Esophageal squamous cell carcinoma (ESCC) is the most common type of esophageal cancer worldwide. The gold standard for its diagnosis requires esophagogastroduodenoscopy, but this is not feasible as a screening modality due to high cost and risk. Therefore, we sought to develop a diagnostic strategy using ESCC-specific methylated DNA biomarkers on esophageal cells obtained via minimally invasive method.

Methods: Using genome-wide methylation data and bioinformatics, we identified candidate ESCC-specific methylated DNA biomarkers, which were initially tested on 48 ESCC-matched normal tissue pairs. These candidate biomarkers were further evaluated in a cross-sectional case-control study of 63 patients from whom esophageal cytology samples were collected using a minimally invasive sponge device. Finally, a multi-biomarker model for ESCC diagnosis was created using a machine learning pipeline with the random subspace method.

Results: Five of the six biomarkers (cg20655070, SLC35F1, TAC1, ZNF132, and ZNF542) exhibited significantly higher methylation levels in tumor DNA vs. matched control tissue DNAs (P < 0.05). When tested on sponge-derived esophageal cytology DNA, these five markers were all significantly hypermethylated in ESCC vs. non-ESCC control DNAs (P < 0.01), and all 5 distinguished ESCCs from controls with areas under receiver operating characteristics curve (AUCs) > 0.76 (P < 0.01). Finally, a discriminatory five-marker-plus-age diagnostic panel was then developed, which demonstrated outstanding biomarker performance (sensitivity = 0.91, specificity = 0.96, and AUC = 0.90).

Discussion: This highly discriminatory biomarker panel is a promising strategy for ESCC diagnosis using low-cost, minimally invasive sampling techniques and merits further study in prospective cancer screening trials.


Disclosures:
Ke Ma: Capsulomics, Inc. – Patent Holder.
Hua-Ling Tsai indicated no relevant financial relationships.
Corey Nolet: Capsulomics, Inc. – Consultant.
Dennis Gong: Capsulomics, Inc. – Employee.
Samson Okello indicated no relevant financial relationships.
Boniface Lumori indicated no relevant financial relationships.
Simran Jit indicated no relevant financial relationships.
Andrew Kalra indicated no relevant financial relationships.
Ludmila Danilova indicated no relevant financial relationships.
Yulan Cheng: Capsulomics, Inc. – Patent Holder.
Stephen Meltzer: Capsulomics, Inc. – Advisory Committee/Board Member, Patent Holder, Stockholder/Ownership Interest (excluding diversified mutual funds).


Ke Ma, MD¹, Hua-Ling Tsai, MSc¹, Corey Nolet, PhD², Dennis Gong, ³, Samson Okello, MD⁴, Boniface Lumori, MD⁴, Simran Jit, MBBS¹, Andrew Kalra, BS¹, Ludmila Danilova, PhD¹, Yulan Cheng, MD¹, Stephen J. Meltzer, MD¹. P1379 - Accurate Detection of Esophageal Squamous Cell Carcinoma (ESCC) Using Machine Learning with Methylated DNA Biomarkers, ACG 2021 Annual Scientific Meeting Abstracts. Las Vegas, Nevada: American College of Gastroenterology.