Domain generalization enables general cancer cell annotation in single-cell and spatial transcriptomics

Sep 21, 2026

Uploaded by Anda

Cancer-Finder is introduced to overcome limits in current malignant cell annotation algorithms for pan-cancer single-cell and spatial transcriptomics data. The deep-learning method uses domain generalization to rapidly identify malignant cells in single-cell datasets, achieving an average accuracy of 95.16%. By leveraging spatial transcriptomic datasets for training replacement, Cancer-Finder identifies malignant spots on spatial slides, demonstrated on clear cell renal cell carcinoma samples, including discovery of a 10-gene signature linked to patient prognosis.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages14

Reading Time35 min

AccessFree to download

Introduction

Malignant cell annotation requirements and challenges

Existing methods for identifying malignant cells and spots

Limitations of current single-cell and spatial approaches

Article [https://doi.org/10.1038/s41467-024-46413-6](https://doi.org/10.1038/s41467-024-46413-6) Domain generalization enables general cancer cell annotation in single-cell and spatial transcriptomics Received: 9 June 2023 | Accepted: 9 February 2024 | | --- | | | | Check for updates | Zhixing Zhong1,2,8, Junchen Hou3,8, Zhixian Yao2,8, Lei Dong 4,8, Feng Liu 5, Junqiu Yue6, Tiantian Wu 3, Junhua Zheng 2, Gaoliang Ouyang 3, Chaoyong Yang 1,2,7 & Jia Song 2 Single-cell and spatial transcriptome sequencing, two recently optimized transcriptome sequencing methods, are increasingly used to study cancer and related diseases. Cell annotation, particularly for malignant cell annotation, is essential and crucial for in-depth analyses in these studies. However, current algorithms lack accuracy and generalization, making it difﬁcult to consistently and rapidly infer malignant cells from pan-cancer data. To address this issue, we present Cancer-Finder, a domain generalization-based deep-learning algorithm that can rapidly identify malignant cells in single-cell data with an average accuracy of 95.16%. More importantly, by replacing the single-cell training data with spatial transcriptomic datasets, Cancer-Finder can accurately identify malignant spots on spatial slides. Applying Cancer-Finder to 5 clear cell renal cell carcinoma spatial transcriptomic samples, Cancer-Finder demonstrates a good ability to identify malignant spots and identiﬁes a gene signature consisting of 10 genes that are signiﬁcantly co-localized and enriched at the tumor-normal interface and have a strong correlation with the prognosis of clear cell renal cell carcinoma patients. In conclusion, CancerFinder is an efﬁcient and extensible tool for malignant cell annotation. It has long been acknowledged that tumor heterogeneity is a signiﬁcant barrier to the development of effective cancer treatments1,2. Single-cell RNA sequencing (scRNA-seq) technology has enabled a comprehensive understanding of intra-and inter-tumor heterogeneity at the level of a single cell, thereby facilitating the development of personalized therapies3. Spatial transcriptomics (ST), which follows in the footsteps of scRNA-seq as a promising sequencing technique, captures the spatial context of transcriptional activity within intact tissue4,5, and is increasingly used in cancer research, resulting in a number of ground-breaking discoveries in the study of cancer heterogeneity6,7. In these studies, the precise annotation of malignant state of single cells or spots (measure units in ST) is essential and fundamental. Currently, malignant cells and spots are identiﬁed primarily through marker genes or copy number variation (CNV) events8. Clustering and marker gene detection make it possible to identify 1Institute of Artiﬁcial Intelligence, Department of Chemical Biology, College of Chemistry and Chemical Engineering, Xiamen University, Xiamen 361102, China. 2Institute of Molecular Medicine, Department of Urology, Renji Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai 200127, China. 3School of Pharmaceutical Sciences, State Key Laboratory of Cellular Stress Biology, School of Life Sciences, Xiamen University, Xiamen 361102, China. 4Department of Pathology, Ruijin Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai 200025, China. 5School of Computing and Information Systems, The University of Melbourne, Carlton, Melbourne, VIC 3053, Australia. 6Department of Pathology, Hubei Cancer Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430030, China. 7Innovation Laboratory for Sciences and Technologies of Energy Materials of Fujian Province (IKKEM), Xiamen 361005, China. 8These authors contributed equally: Zhixing Zhong, Junchen Hou, Zhixian Yao, Lei Dong. e-mail: [songjiajia2010@shsmu.edu.cn](songjiajia2010@shsmu.edu.cn) malignant cells/spots, but technical artifacts such as drop-out and high sparsity can lea

FAQs

What problem does Cancer-Finder address?

Current algorithms struggle with accuracy and generalization for consistently and rapidly inferring malignant cells from pan-cancer single-cell data.

How does Cancer-Finder work?

Cancer-Finder is a domain generalization-based deep-learning approach designed to rapidly identify malignant cells and can be adapted by replacing single-cell training data with spatial transcriptomic datasets.

What evidence supports Cancer-Finder’s performance and usefulness?

The method reports an average single-cell identification accuracy of 95.16% and shows the ability to identify malignant spots in spatial slides, including a 10-gene signature associated with clear cell renal cell carcinoma prognosis.

Sep 21, 2026

Uploaded by Anda