Related Experiment Video
Updated: Jan 11, 2026
09:49
Oncogenic Gene Fusion Detection Using Anchored Multiplex Polymerase Chain Reaction Followed by Next Generation Sequencing
Published on: July 5, 2019
9.9K
Early cancer detection via multi-omics cfDNA fragmentation using early-late fusion neural network with
Libo Lu1,2,3, Yunze Wang1,2,3, Xionghui Zhou1,2,3
1Hubei Key Laboratory of Agricultural Bioinformatics, College of Informatics, Huazhong Agricultural University, No.1 Shizishan Street, Hongshan District, Wuhan, Hubei 430070, People's Republic of China.
Briefings in Bioinformatics
|November 9, 2025
Summary
Cell-free DNA (cfDNA) fragmentation patterns show promise for early cancer detection. A new method, ELSM, effectively integrates multiple cfDNA signals to improve cancer diagnosis and identify tissue of origin.
Area of Science:
- Biochemistry
- Genomics
- Computational Biology
Background:
- Cell-free DNA (cfDNA) fragmentation patterns are linked to epigenetic modifications, making them potential biomarkers for early cancer detection.
- Integrating diverse fragmentomic signals can enhance diagnostic accuracy, but challenges like high dimensionality and limited samples hinder multimodal fusion.
Purpose of the Study:
- To develop and validate a novel deep learning framework, Early-Late fusion with Sample-Modality evaluation (ELSM), for improved cancer detection and tissue-of-origin prediction using cfDNA fragmentation patterns.
- To assess ELSM's performance against existing unimodal and multimodal approaches across multiple cancer datasets.
Main Methods:
- ELSM, a two-stage neural network, integrates 13 fragmentomic feature spaces.
- Incorporates sample-wise modality evaluation to capture complementary signals effectively.
- Validated across five datasets comprising 1994 samples from 10 cancer types.
Main Results:
- ELSM achieved an Area Under the Curve (AUC) of 0.972 for pan-cancer diagnosis.
- Demonstrated strong performance in an independent gastric cancer cohort (AUC 0.922).
- Achieved a median tissue-of-origin prediction accuracy of 0.683, outperforming existing models.
Conclusions:
- ELSM offers a powerful and interpretable framework for integrative multi-omics analysis.
- The method effectively captures complementary cfDNA fragmentomic signals for enhanced cancer detection.
- ELSM shows significant potential for clinical translation in early cancer diagnostics.
