Related Experiment Video
Updated: Jul 16, 2026
Discovery of Driver Genes in Colorectal HT29-derived Cancer Stem-Like Tumorspheres
Published on: July 22, 2020
AI-driven multi-omics deciphering of extrachromosomal circular DNA rescues hidden targets in colorectal cancer
Lili Zhang1, Jian Cui2, Tianhan Sun2
1Clinical Biobank, Beijing Hospital, National Center for Gerontology; National Clinical Research Center for Gerontology; The Key Laboratory of Geriatrics of NHC; Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, Beijing, People's Republic of China.
Background:
Extrachromosomal circular DNA (eccDNA) is an emerging tumor biomarker notable for its tumor-specific amplification and contribution to genomic heterogeneity. However, the substantial heterogeneity of eccDNAs poses a significant challenge for characterization. Current identification tools rely predominantly on single data types, overlooking the integrative potential of multi-omics layers and often discarding sparse but biologically meaningful signals.
Methods:
To address this, we developed eccDNAOmix, an AI-empowered framework that integrates a specialized eccDNA sequencing pipeline with multi-omics data to evaluate the relative contributions of different biological modalities. Basing matched tumor and adjacent normal tissues from colorectal cancer patients, eccDNAOmix employs a masking strategy to eliminate non-biological noise and features a multimodal deep learning model with an adaptive gated fusion mechanism.
Results:
Our model achieved robust identification performance (AUC = 0.844), with the DNA sequence modality (AUC = 0.822) being the most identifiable feature for eccDNA identification. Leveraging this finding, we established a web server to facilitate practical eccDNA identification. Biological characterization showed that eccDNAs are preferentially derived from promoter regions and exhibit a significant enrichment of RNA modifications within 100 bp downstream of junction sites. Notably, eccDNAOmix rescued hidden eccDNA fragments from conventionally discarded unmapped sequencing reads and this was confirmed via experimental validation.
Conclusions:
By integrating multi-omics sequencing and eccDNA sequencing, this study provides the first comprehensive characterization of eccDNAs. Our multi-omics analysis demonstrates that the DNA sequence contributes the highest proportion to eccDNA identification. By leveraging this dominant identifiable role and providing an accessible single-modality identification tool, our work proposes a practical computational framework for advancing discovery in colorectal cancer.
Insights
Extrachromosomal circular DNA (eccDNA) characterization is challenging due to heterogeneity. Our AI framework, eccDNAOmix, integrates multi-omics data, identifying DNA sequence as the key feature for eccDNA discovery in colorectal cancer.
Area of Science:
- Genomics
- Bioinformatics
- Cancer Research
Background:
- Extrachromosomal circular DNA (eccDNA) are emerging tumor biomarkers with significant heterogeneity, posing challenges for characterization.
- Current identification methods often rely on single data types, overlooking multi-omics potential and discarding sparse signals.
Purpose of the Study:
- To develop an AI-powered framework, eccDNAOmix, for comprehensive eccDNA characterization by integrating multi-omics data.
- To evaluate the relative contributions of different biological modalities in eccDNA identification.
Main Methods:
- Developed eccDNAOmix, an AI framework integrating eccDNA sequencing with multi-omics data from colorectal cancer tissues.
- Employed a masking strategy to reduce noise and a multimodal deep learning model with adaptive gated fusion.
Main Results:
- Achieved robust eccDNA identification performance (AUC=0.844), with DNA sequence being the most identifiable modality (AUC=0.822).
- Established a web server for practical eccDNA identification and identified preferential derivation from promoter regions with enriched RNA modifications.
- Recovered previously discarded eccDNA fragments from unmapped sequencing reads, validated experimentally.
Conclusions:
- This study provides the first comprehensive characterization of eccDNAs by integrating multi-omics and eccDNA sequencing.
- DNA sequence is the dominant feature for eccDNA identification.
- Proposes a practical computational framework for advancing eccDNA discovery in colorectal cancer.
