Open imaging and AI resource enables large-scale quantification of extrachromosomal DNA in cancer

A preprint posted to bioRxiv describes an open computational resource designed to accelerate research into extrachromosomal DNA (ecDNA) — circular DNA elements that exist outside normal chromosomes and are found in a variety of cancer types. ecDNA is of growing research interest because it can carry oncogene copies at high amplification levels, contributes to intratumour heterogeneity, and has been associated with poor prognosis and resistance to treatment in several cancer contexts.

The resource integrates 2,986 native-resolution metaphase FISH (fluorescence in situ hybridisation) image sets with expert manual annotations, standardised benchmarks, and open-source computational frameworks for automated ecDNA detection and quantification. The authors use this dataset to systematically compare existing and newly developed analytical approaches, spanning rule-based algorithms and machine-learning methods, identifying strengths and limitations of each.

Prior to this work, automated ecDNA analysis has been constrained by the absence of accessible, annotated imaging data and validated computational tools. By releasing both data and code openly, the authors aim to enable reproducible, unbiased quantification of ecDNA across research groups. The preprint has not yet been peer reviewed. Researchers working on cancer genomics, structural variation, or intratumour heterogeneity may find the dataset and benchmarks directly applicable to their own studies.