Tracing unknown tumor origins - A biological-pathway-based transformer model

Sep 21, 2026

Uploaded by Violet

Xie and colleagues develop BPformer, a deep learning tool that couples a transformer-encoder model with biological pathway prior knowledge to identify tumor origin in cancers of unknown primary (CUP). The method transforms high-dimensional gene expression profiles into an analyzable representation and improves interpretability of oncogenic mechanisms driving metastasis. Trained on transcriptomes from 10,410 primary tumors across 32 cancer types, BPformer achieves high primary and metastatic-site accuracy and is further validated in a retrospective cohort.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Motivation

Summary

Introduction

Table 1. Methods in transcriptome data of cancer origin tracing

Article Tracing unknown tumor origins with a biologicalpathway-based transformer model Authors Jiajing Xie, Ying Chen, Shijie Luo, ..., Xin Ding, Mengsha Tong, Rongshan Yu Correspondence [xinding2014@gmail.com](xinding2014@gmail.com) (X.D.), [mstong@xmu.edu.cn](mstong@xmu.edu.cn) (M.T.), [rsyu@xmu.edu.cn](rsyu@xmu.edu.cn) (R.Y.) In brief Xie et al. develop BPformer, a deep learning tool that integrates a transformer model with biological pathways to accurately identify tumor origin in cancers of unknown primary (CUPs) . The tool provides essential insights into oncogenic pathways and improves diagnostic processes through an accessible web server for researchers and clinicians worldwide. d BPformer diagnoses the origin in cancer of unknown primary (CUPs) d Incorporating pathways into transformer model improves predictive accuracy d BPformer identiﬁes key pathways and uncovers new mechanisms of cancer progression d A user-friendly web server for CUP diagnostics facilitates clinical use Xie et al., 2024, Cell Reports Methods 4, 100797 June 17, 2024 ª 2024 The Authors. Published by Elsevier Inc. [https://doi.org/10.1016/j.crmeth.2024.100797](https://doi.org/10.1016/j.crmeth.2024.100797) ll ll OPEN ACCESS Article Tracing unknown tumor origins with a biological-pathway-based transformer model Jiajing Xie,1,6 Ying Chen,2,6 Shijie Luo,1,6 Wenxian Yang,3 Yuxiang Lin,1 Liansheng Wang,1,2 Xin Ding,4,* Mengsha Tong,1,5,* and Rongshan Yu1,2,3,7,* 1National Institute for Data Science in Health and Medicine, Xiamen University, Xiamen, Fujian 361102, China 2School of Informatics, Xiamen University, Xiamen, Fujian 361005, China 3Aginome Scientiﬁc, Xiamen, Fujian 361005, China 4Department of Pathology, Zhongshan Hospital of Xiamen University, School of Medicine, Xiamen University, Xiamen, Fujian 361004, China 5State Key Laboratory of Cellular Stress Biology, School of Life Sciences, Faculty of Medicine and Life Sciences, Xiamen University, Xiamen, Fujian 361102, China 6These authors contributed equally 7Lead contact *[Correspondence:](Correspondence: xinding2014@gmail.com)[ xinding2014@gmail.com](Correspondence: xinding2014@gmail.com) (X.D.), [mstong@xmu.edu.cn](mstong@xmu.edu.cn) (M.T.), [rsyu@xmu.edu.cn](rsyu@xmu.edu.cn) (R.Y.) [https://doi.org/10.1016/j.crmeth.2024.100797](https://doi.org/10.1016/j.crmeth.2024.100797) MOTIVATION Gene expression proﬁles contain rich information about the primary site of tumors, but directly processing these high-dimensional data for tumor origin identiﬁcation poses signiﬁcant challenges. To tackle this challenge, we developed BPformer, a transformer-encoder-based model to incorporate prior knowledge in biological pathways to integrate high-dimensional expression data into a manageable dimension for analysis. BPformer demonstrates superior performance in tracing the origins of cancer and improves the interpretability of the mechanisms underlying tumor metastasis, making ita valuable tool for cancer research. SUMMARY Cancer of unknown primary (CUP) represents metastatic cancer where the primary site remains unidentiﬁed despite standard diagnostic procedures. To determine the tumor origin in such cases, we developed BPformer, a deep learning method integrating the transformer model with prior knowledge of biological pathways. Trained on transcriptomes from 10,410 primary tumors across 32 cancer types, BPformer achieved remarkable accuracy rates of 94%, 92%, and 89% in primary tumorsand primary and metastatic sites of metastatic tumors, respectively, surpassing existing methods. Additionally, BPformer was validated in a retrospective study, demonstrating consistency with tumor sites diagnosed through immunohistochemistry and histopathology. Furthermore, BPformer was able to rank pathways based on their contribution to tumor origin identiﬁcation, which helped to classify oncogenic signaling pathways into those that are highly conservative among di

FAQs

What problem does BPformer address?

BPformer addresses cancer of unknown primary (CUP), where the primary tumor site cannot be identified through standard diagnostic procedures.

How does BPformer improve tumor origin identification?

BPformer integrates biological pathway prior knowledge into a transformer-encoder framework to incorporate interpretable biological information and improve predictive accuracy.

What data and validation support BPformer’s performance?

BPformer is trained on transcriptomes from 10,410 primary tumors across 32 cancer types and is validated in a retrospective study with consistency to tumor sites diagnosed by immunohistochemistry and histopathology.

Sep 21, 2026

Uploaded by Violet