Paper Outlook — 2026 年 8 月 31 日至 9 月 6 日

· 3 min read · Paper Feed

Paper Outlook — 2026 年 8 月 31 日至 9 月 6 日

摘要

  • 空间转录组学在快速走向成熟：SpaGCN 用图神经网络把基因表达、空间位置和组织学信息拼到同一张图里；BayesSpace 走的是贝叶斯邻域模型的路子，靠先验把分辨率往高了拉；G2T 思路反向，仅凭基因表达，借助生成式流匹配去预测组织原本的空间布局。
  • 单细胞和单核 RNA 测序越来越多被用来回答具体的生物学问题，不再只是发新工具：比如肥胖怎样重塑子宫内膜样腺癌里的雌激素信号，又比如中药小檗碱是不是通过影响线粒体自噬（细胞清理受损线粒体的过程）来抑制乳腺癌。
  • 学界开始怀疑靠人工标记基因给细胞分型的做法是否真够用。GmGM 框架反过来，把细胞和基因之间的关系一并学习，据说能抓到常规聚类里被合并掉的那些细微细胞状态。
  • 「多组学 + 湿实验验证」越来越像药物机制研究的标配流程：公共数据集、单细胞数据、分子对接一起上，实验台再补一刀，用来筛靶点、做验证。
  • 生成式深度学习慢慢渗进空间生物学，预示着一类转变正在发生：从对组织结构做纯粹的描述，往能预测、能补齐的方向走——尤其在空间数据拿不到或者代价太大的场景里。

单核 RNA 测序揭示肥胖相关子宫内膜样腺癌中雌激素信号通路的重编程。

Single-nuclei characterization of endometrioid adenocarcinoma (A) UMAP of 42,574 nuclei from 13 primary tumors, colored and labeled by six major cell types: ciliated epithelial, endothelial, lymphoid, myeloid, stromal/myometrial, and tumor epithelial. (B) Expression of canonical marker genes for cluster annotation: DNAH12 (ciliated epithelium), VWF (endothelium), IKZF1 (lymphoid), MS4A6A (myeloid), COL3A1 (stromal/myometrial), and GRIA2 (tumor epithelial). (C) UMAPs visualizing the integrated dataset colored by clinical and pathological variables (race, patient of origin, tumor grade, BMI) to assess potential batch effects. (D) Composition of each tumor sample by cluster. (E) Distribution of body mass index group across each identified cluster. n = 13 tumors (7 obese, 6 normal-weight), 42,574 nuclei total. No formal statistical tests applied. See also Figures S1 , S2 , S3 , S4 , and S5 .

Figure 1，出自 Long JL 等, iScience (2026)。许可协议 CC BY-NC-ND。来源：10.1016/j.isci.2026.117226。

iScience (2026)

阅读原文 → https://doi.org/10.1016/j.isci.2026.117226

小檗碱调控线粒体自噬相关基因表达并抑制乳腺癌细胞：一项多组学与实验研究。

概念插图：Berberine Modulates Mitophagy-Related Gene Expression and Suppresses Breast Cancer Cells: A Multi-Omics and Experimental Study.

Cell biochemistry and biophysics (2026)

阅读原文 → https://doi.org/10.1007/s12013-026-02156-3

SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network

概念插图：SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network

Nature Methods (2021) · 1024 次引用

阅读原文 → https://doi.org/10.1038/s41592-021-01255-8

基于 BayesSpace 实现亚spot 级别的空间转录组学分析

概念插图：Spatial transcriptomics at subspot resolution with BayesSpace

Nature Biotechnology (2021) · 892 次引用

阅读原文 → https://doi.org/10.1038/s41587-021-00935-2

G2T: Tissue Reconstruction from Gene Expression via Embedding-Distance Flow Matching

概念插图：G2T: Tissue Reconstruction from Gene Expression via Embedding-Distance Flow Matching

bioRxiv (2026)

阅读原文 → https://doi.org/10.64898/2026.08.25.746917

通过无偏细胞与基因网络分析揭示 scRNAseq 中的细胞分辨率

概念插图：Uncovering Cellular Resolution in scRNAseq via Unbiased Cell and Gene Network Analysis

arXiv (2026)

阅读原文 → https://arxiv.org/abs/2608.22982v1

快速链接

论文 期刊 / 年份 链接
Single-nuclei RNA sequencing reveals obesity-associated rewiring of estrogen signaling in endometrioid adenocarcinoma. iScience (2026) 阅读 →
Berberine Modulates Mitophagy-Related Gene Expression and Suppresses Breast Cancer Cells: A Multi-Omics and Experimental Study. Cell biochemistry and biophysics (2026) 阅读 →
SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network Nature Methods (2021) 阅读 →
Spatial transcriptomics at subspot resolution with BayesSpace Nature Biotechnology (2021) 阅读 →
G2T: Tissue Reconstruction from Gene Expression via Embedding-Distance Flow Matching bioRxiv (2026) 阅读 →
Uncovering Cellular Resolution in scRNAseq via Unbiased Cell and Gene Network Analysis arXiv (2026) 阅读 →

Cancer Genomics Oncology Tumor Spatial Transcriptomics Spatial Omics Single-Cell Transcriptomics