Related Experiment Video
Updated: Jul 12, 2026
11:11
Detection of Rare Mutations in CtDNA Using Next Generation Sequencing
Published on: August 24, 2017
A scalable deep-learning framework for cancer detection using cell-free DNA shallow whole-genome sequencing
Haichao Wang1,2,3, Paulius D Mennea1,2,3, Grainne McAndrew1,2,3
1Centre for Cancer Cell and Molecular Biology, Barts Cancer Institute, Queen Mary University of London, John Vane Science Centre, Charterhouse Square, London EC1M 6BQ, UK.
Science Advances
|July 10, 2026
Summary
This study introduces UNITE, a novel AI framework for noninvasive cancer detection using cell-free DNA (cfDNA). UNITE enhances early cancer screening sensitivity by analyzing fragmentomic and epigenetic features from shallow whole-genome sequencing data.
Area of Science:
- Biomarkers
- Genomics
- Artificial Intelligence
Background:
- Cell-free DNA (cfDNA) in body fluids offers potential for noninvasive cancer detection.
- Multifeature artificial intelligence (AI) can improve cancer detection sensitivity by integrating diverse biomarkers, especially when cancer signals are sparse.
- Tumor-naive approaches utilizing fragmentomic and epigenetic features are emerging for screening individuals with low tumor burden, overcoming limitations of mutation-based assays for early detection.
Purpose of the Study:
- To design UNITE, a universal cfDNA feature ensemble framework for scalable cancer detection.
- To evaluate AI models (XGBoost and CNN) within the UNITE framework using shallow whole-genome sequencing (sWGS) data.
- To assess the performance of UNITE in detecting various cancer types across different stages.
Main Methods:
- Developed UNITE, a framework generating "genomic bin-fragment length" matrices from 0.1× depth sWGS data.
- Utilized plasma samples from 2063 individuals (631 controls, 1432 cancer cases across 26 types).
- Systematically evaluated XGBoost (UNITE-XGB) and convolutional neural networks (UNITE-CNN) across multiple feature spaces and cancer stages.
Main Results:
- UNITE-XGB achieved 31% sensitivity and UNITE-CNN achieved 21% sensitivity in stage I-II cancer detection.
- Both models maintained 95% specificity in early-stage cancer detection.
- The study demonstrated the feasibility of using cfDNA fragmentomic and epigenetic features for early cancer screening.
Conclusions:
- UNITE provides a scalable, multifeature AI approach for noninvasive cancer detection using cfDNA.
- The findings offer a roadmap for developing advanced AI-driven cancer screening methods beyond traditional plasma biopsies.
- This framework holds promise for improving early cancer detection rates, particularly in individuals with low tumor burden.
