Scholarlink

Background <p>Tumor microenvironment (TME) plays a crucial role in cancer progression, metastasis, and treatment response. Recent advances in single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics (ST) have provided valuable insights into the cellular diversity and spatial organization of the TME. However, prioritizing clinically relevant cellular subpopulations in spatial contexts using high-dimensional and sparse data remains a challenge.</p> Methods <p>We introduce TiRank, a novel framework designed to prioritize clinically relevant spatial niches. TiRank incorporates a relative expression ordering (REO)-transformation module to mitigate systematic biases across modalities and utilizes a multitask transfer learning framework to align scRNA-seq, ST, and bulk transcriptomes into a unified embedding space. We benchmarked TiRank using multiple public datasets and pan-cancer clinical cohorts, demonstrating its capability to identify phenotypic cell subpopulations and spatial niches.</p> Results <p>By integrating scRNA-seq, ST, and bulk transcriptomics with clinical phenotypes, TiRank demonstrates high accuracy in identifying drug-sensitive cells and clinically relevant spatial niches across various cancer types. As a case study, we applied TiRank to gastric cancer (GC) to prioritize spatial niches associated with patient outcomes. In our clinical cohort, TiRank successfully revealed a distinct spatial niche at the tumor boundary, characterized by an enrichment of cancer-associated fibroblasts (CAFs). This niche was associated with the efficacy of different treatment regimens. To validate this finding, we further performed multiplex protein imaging on an independent cohort to confirm the spatial distribution of the CAFs-enriched barrier. Moreover, this barrier, termed Fibro-Bar, was strongly correlated with treatment response to neo-adjuvant chemoimmunotherapy. To improve accessibility, we developed TiRank as an open-source tool with an interactive graphical user interface for both researchers and clinicians.</p> Conclusions <p>TiRank offers a phenotype-guided, cross-modal strategy to prioritize clinically relevant spatial niches by coupling an REO-based representation with transfer learning from bulk clinical cohorts. This design enables clinically supervised niche prioritization without requiring large, matched single-cell or spatial clinical cohorts, advancing biomarker discovery and supporting precision oncology.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

TiRank prioritizes phenotypic niches in tumor microenvironment for clinical biomarker discovery

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月13日

查看全文
  • Yuxiang Lin,
  • Zening Huang,
  • Ziyan Lin,
  • Yating Lin,
  • Jinsheng Song,
  • Ling Luo,
  • Jiayao Chi,
  • Yeyang Zheng,
  • Youxin Gao,
  • Junjie Lin,
  • Xinyu Li,
  • Chenyu Liang,
  • Lei Zhang,
  • Xinkang Wang,
  • Yuqin Sun,
  • Rongshan Yu,
  • Qiyue Chen,
  • Mengsha Tong

摘要

Background

Tumor microenvironment (TME) plays a crucial role in cancer progression, metastasis, and treatment response. Recent advances in single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics (ST) have provided valuable insights into the cellular diversity and spatial organization of the TME. However, prioritizing clinically relevant cellular subpopulations in spatial contexts using high-dimensional and sparse data remains a challenge.

Methods

We introduce TiRank, a novel framework designed to prioritize clinically relevant spatial niches. TiRank incorporates a relative expression ordering (REO)-transformation module to mitigate systematic biases across modalities and utilizes a multitask transfer learning framework to align scRNA-seq, ST, and bulk transcriptomes into a unified embedding space. We benchmarked TiRank using multiple public datasets and pan-cancer clinical cohorts, demonstrating its capability to identify phenotypic cell subpopulations and spatial niches.

Results

By integrating scRNA-seq, ST, and bulk transcriptomics with clinical phenotypes, TiRank demonstrates high accuracy in identifying drug-sensitive cells and clinically relevant spatial niches across various cancer types. As a case study, we applied TiRank to gastric cancer (GC) to prioritize spatial niches associated with patient outcomes. In our clinical cohort, TiRank successfully revealed a distinct spatial niche at the tumor boundary, characterized by an enrichment of cancer-associated fibroblasts (CAFs). This niche was associated with the efficacy of different treatment regimens. To validate this finding, we further performed multiplex protein imaging on an independent cohort to confirm the spatial distribution of the CAFs-enriched barrier. Moreover, this barrier, termed Fibro-Bar, was strongly correlated with treatment response to neo-adjuvant chemoimmunotherapy. To improve accessibility, we developed TiRank as an open-source tool with an interactive graphical user interface for both researchers and clinicians.

Conclusions

TiRank offers a phenotype-guided, cross-modal strategy to prioritize clinically relevant spatial niches by coupling an REO-based representation with transfer learning from bulk clinical cohorts. This design enables clinically supervised niche prioritization without requiring large, matched single-cell or spatial clinical cohorts, advancing biomarker discovery and supporting precision oncology.