[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-346014-105":3,"doc-detail-346014-en":81,"detail-sidebar-cat-0-en-105":115},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","decoding-cancer-circulating-transcriptomic-signatures-with-language-models","Decoding cancer circulating transcriptomic signatures with language models","","Sequence-level modelling of plasma cell-free RNA (cfRNA) enables cancer detection beyond gene annotations. The work introduces GeneLLM, a Transformer-based model that directly processes nucleotide sequences from human-mapped cfRNA reads to learn latent pseudo-biomarkers as discriminative features. In a multi-centre cohort, GeneLLM attains ROC-AUC values spanning 0.9250–0.9962 across several cancers and preserves comparable performance at one-sixth of typical sequencing depth, supporting more cost-efficient, scalable screening. It captures signals from transcriptomic dark matter.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/decoding-cancer-circulating-transcriptomic-signatures-with-language-models/346014/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/decoding-cancer-circulating-transcriptomic-signatures-with-language-models/346014.png","ImageObject",300,407,{"name":42,"@type":43},"Clementine","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-17","2026-09-22",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",8,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does GeneLLM address in multi-cancer liquid biopsy?","Question",{"text":63,"@type":64},"Current cfRNA liquid biopsy methods often depend on gene annotations, which can miss signals from unannotated or repetitive genomic regions. GeneLLM instead processes nucleotide sequences directly to identify cancer-indicative signatures.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How does GeneLLM generate cancer-related features without gene-level quantification?",{"text":68,"@type":64},"GeneLLM learns latent pseudo-biomarkers—prototype representations derived from aggregated cfRNA read embeddings—used as discriminative features for cancer classification rather than mapped genomic sequences.",{"name":70,"@type":61,"acceptedAnswer":71},"What performance and cost-related results are reported for GeneLLM?",{"text":72,"@type":64},"In a multi-centre cohort, GeneLLM achieves ROC-AUC values from 0.9250 to 0.9962 across several cancers and maintains comparable performance at one-sixth of the typical sequencing depth, supporting more cost-efficient screening.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},346014,1790960149,{"code":4,"msg":82,"data":83},"success",{"doc_id":79,"user_id":84,"nickname":42,"user_avatar":85,"doc_module":4,"category_id":55,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":86,"file_id":87,"file_url":88,"file_type":89,"file_size":90,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":91,"language":92,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":93,"faqs":94,"seo_title":95,"seo_description":12,"update_tm":96,"read_time":97,"preview_page_count":33,"preview_urls":98},1374391974564,"https://ap-avatar.wpscdn.com/avatar/14000253aa45c000a9e?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779874745381141002","Article [https://doi.org/10.1038/s41467-026-74411-3](https://doi.org/10.1038/s41467-026-74411-3)  \nDecoding cancer circulating transcriptomic signatures with language models  \nReceived: 29 April 2025  \n\n| Accepted: 3 June 2026 |\n| --- |\n| |\n| Check for updates |\n\nSiwei Deng 1,2,3,4,10, Lei Sha 2,3,4,5,10 , Yongcheng Jin3,4,10 ,  \nTianyao Zhou3,4,10, Chengen Wang6,10, Qianpu Liu2, Hongjie Guo7,  \nChengjie Xiong8, Yangtao Xue8, Xiaoguang Li 6, Yuanming Li6, Yaping Gao3, Mengyu Hong9, Junjie Xu 8 , Shanwen Chen 1  & Pengyuan Wang 1   \nCurrent liquid biopsy methods for multi-cancer detection using plasma cellfree RNA (cfRNA, short RNA fragments circulating in blood that can reﬂect disease states) typically rely on gene annotations, which can overlook signals from unannotated or repetitive genomic regions. We present GeneLLM, a Transformer-based model that directly processes the nucleotide sequences of human-mapped cfRNA reads to identify cancer-indicative signatures. By bypassing gene-level quantiﬁcation, the model retains signals from transcriptomic dark matter. The model learns latent pseudo-biomarkers (prototype representations from aggregated cfRNA read embeddings) that serve as discriminative features for cancer classiﬁcation, rather than corresponding to explicit genomic sequences. Here we show that, in a multi-centre cohort, GeneLLM achieves ROC-AUC values ranging from 0.9250 to 0.9962 across several cancers, while maintaining comparable performance at one-sixth of the typical sequencing depth. These results suggest that sequence-level modelling of plasma cfRNA can capture diagnostically relevant information beyond annotation-dependent approaches, enabling more cost-efﬁcient and scalable cancer screening.  \nEarly cancer detection is crucial for improving treatment efﬁcacyand patient outcomes1,2. Non-invasive, cost-effective techniques, such as those leveraging cell-free RNA (cfRNA), are increasingly important for this purpose. CfRNA provides advantages over cellfree DNA (cfDNA) for solid tumour detection by reﬂecting dynamic gene expression proﬁles in response to disease3–9. This approach addresses limitations of existing methods like  \ncolonoscopy and mammography, which can be invasive, expensive, and resource-intensive10,11.  \nNonetheless, cfRNA-based detection encounters challenges of sensitivity and speciﬁcity, arising from ﬂuctuating cfRNA levels and thedifﬁculty in differentiating cancerous and normal proﬁles. To address this, studies have integrated machine learning methods, performing differential expression (DE) analysis to discover differentially  \n1Division of Gastrointestinal Surgery, Peking University First Hospital, Peking University, Beijing, China. 2Institute of Artiﬁcial Intelligence, Beihang University, Beijing, China. 3OxTium Technology Co., Ltd, Shenzhen, China. 4Research Institute of Tsinghua University in Shenzhen, Building A, No. 39 Gaoxin South 9th Road, Nanshan District, Shenzhen, China. 5Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, Beihang University, Beijing, China. 6Department of Minimally Invasive Tumor Therapies Center, Beijing Hospital, National Center of Gerontology, Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, Beijing, China. 7Department of Interventional Radiology and Vascular Surgery, Peking University First Hospital, Peking University, Beijing, China. 8Department of General Surgery, Sir Run-Run Shaw Hospital, Zhejiang University School of Medicine, Hangzhou, China. 9Academy for Advanced Interdisciplinary Science and Technology, Key Laboratory of Advanced Materials and Devices for Post-Moore Chips Ministry of Education, University of Science and Technology Beijing, Beijing, China. 10These authors contributed equally: Siwei Deng, Lei Sha, Yongcheng Jin, Tianyao Zhou,  \nChengen Wang. e-mail: [lei.sha@oxtium.com](lei.sha@oxtium.com); [yongcheng.jin@oxtium.com](yongcheng.jin@oxtium.com); [walter235@zju.edu.cn](walter235@zju.edu.cn); [","cbCaiayqqohtKQJP","https://ap.wps.com/l/cbCaiayqqohtKQJP","pdf",4624107,16,"English","# Background\n## Need for early, non-invasive cancer detection\n## Limitations of cfRNA sensitivity/specificity\n# Method\n## GeneLLM sequence-level model\n## Latent pseudo-biomarkers\n# Results\n## Multi-centre performance and ROC-AUC\n## Robustness under reduced sequencing depth\n# Implications\n## Annotation-independent cancer screening","[{\"question\":\"What problem does GeneLLM address in multi-cancer liquid biopsy?\",\"answer\":\"Current cfRNA liquid biopsy methods often depend on gene annotations, which can miss signals from unannotated or repetitive genomic regions. GeneLLM instead processes nucleotide sequences directly to identify cancer-indicative signatures.\"},{\"question\":\"How does GeneLLM generate cancer-related features without gene-level quantification?\",\"answer\":\"GeneLLM learns latent pseudo-biomarkers—prototype representations derived from aggregated cfRNA read embeddings—used as discriminative features for cancer classification rather than mapped genomic sequences.\"},{\"question\":\"What performance and cost-related results are reported for GeneLLM?\",\"answer\":\"In a multi-centre cohort, GeneLLM achieves ROC-AUC values from 0.9250 to 0.9962 across several cancers and maintains comparable performance at one-sixth of the typical sequencing depth, supporting more cost-efficient screening.\"}]","Decoding cancer circulating transcriptomic signatures with language models | PDF",1790059688,40,[99,103,107,111],{"page_no":22,"url":100,"width":101,"height":102},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/ea9c05a86763813e3ee00b89f490af39.webp",983,1305,{"page_no":26,"url":104,"width":105,"height":106},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/7421c9fd94129a2cd96e22db5930a2da.webp",982,1304,{"page_no":30,"url":108,"width":109,"height":110},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/d82bd47e4bcb02267bd2ce4d2591e14f.webp",981,1303,{"page_no":33,"url":112,"width":113,"height":114},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/9b5e756cb889343f4a63309552491963.webp",980,1301,{"code":4,"msg":82,"data":116},[117,121,125,129,134,139,143,146,151,154,158],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":118,"show_sort_weight":119,"slug":120},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":122,"show_sort_weight":123,"slug":124},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":126,"show_sort_weight":127,"slug":128},"Exam",70,"exam",{"id":130,"doc_module":4,"doc_module_name":25,"category_name":131,"show_sort_weight":132,"slug":133},5,"Comic",60,"comic",{"id":135,"doc_module":4,"doc_module_name":25,"category_name":136,"show_sort_weight":137,"slug":138},6,"Technology",50,"technology",{"id":140,"doc_module":4,"doc_module_name":25,"category_name":141,"show_sort_weight":97,"slug":142},7,"Healthcare","healthcare",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":144,"slug":145},30,"research-report",{"id":147,"doc_module":4,"doc_module_name":25,"category_name":148,"show_sort_weight":149,"slug":150},9,"Religion & Spirituality",20,"religion-spirituality",{"id":149,"doc_module":4,"doc_module_name":25,"category_name":152,"show_sort_weight":149,"slug":153},"World Cup","world-cup",{"id":155,"doc_module":4,"doc_module_name":25,"category_name":156,"show_sort_weight":155,"slug":157},10,"Lifestyle","lifestyle",{"id":159,"doc_module":4,"doc_module_name":25,"category_name":160,"show_sort_weight":130,"slug":161},19,"General","general"]