Paper Outlook — August 24-30, 2026

· 5 min read · Paper Feed

Paper Outlook — August 24-30, 2026

TL;DR

  • Deep learning and statistical learning tools are advancing spatial transcriptomics — from 3D tissue reconstruction to mapping how cells organize within tumors.
  • New computational frameworks are inferring gene regulatory networks from single-cell multi-omics data, even when measurements are sparse and noisy.
  • Several studies use single-cell and spatial omics to pinpoint cancer-driving genes and metabolic biomarkers, with colorectal cancer as a recurring focus.
  • Distinguishing malignant cells from surrounding normal cells at single-cell resolution remains a common challenge, and new methods keep tackling it.

SPACEL: deep learning-based characterization of spatial transcriptome architectures

SPACEL is a deep learning toolkit for analyzing spatial gene expression data. It predicts what cell types sit at each sampled spot, aligns multiple tissue slices from the same sample, and stacks those slices into a 3D model of the tissue. Most spatial experiments currently give you 2D pictures, so being able to rebuild a full 3D tissue architecture opens new ways to study how cells organize themselves in health and disease.

Workflow of SPACEL. a Three modules of SPACEL: Spoint, Splane, and Scube. b Spoint deconvolutes cell types of spots using an MLP and a probability model. MLP, multiple-layer perceptron; VAE, variational autoencoder. c Splane employs a GCN model and an adversarial learning algorithm to identify spatial domains across multiple slices. GCN, graph convolutional network. d For consecutive slices, Scube adopts a mutual nearest neighbor (MNN) graph and the differential evolution algorithm to transform slices and construct a stacked 3D alignment of a tissue.

Fig. 1 from Hao Xu et al., Nature Communications (2023). Licensed CC BY. Source: 10.1038/s41467-023-43220-3.

Nature Communications (2023) · 114 citations

Read the original → https://doi.org/10.1038/s41467-023-43220-3

SpaTopic: A statistical learning framework for exploring tumor spatial architecture from spatially resolved transcriptomic data

SpaTopic borrows a 'topic modeling' idea from text mining to find recurring spatial neighborhoods inside tumor tissue. It combines spot clustering with cell-type deconvolution from single-cell data, producing spatial domains that read more clearly in biological terms. The end result is a sharper picture of the tumor microenvironment and how its cellular neighborhoods map onto disease features.

Schematic overview of SpaTopic. ( A ) SpaTopic is designed to infer, annotate, and compare spatial domains based on single-cell transcriptomic (SCT) data with cell-type annotation and SRT data. In the modeling phase, spatial clustering and deconvolution analyses are performed based on the input SCT and SRT data. Then, a cell type–specific score matrix is calculated to reflect the significance of enrichment of cell types in different spatial clusters. Subsequently, topic modeling with LDA is applied to the specific score matrix to extract cell-type topics across different spot clusters. The cluster-topic matrix is binarized to refine the initial spot clusters into spatial domains, which show unique distribution of cell-type compositions (CellTopics). CellTopics from different SRT datasets can be quantitatively compare, in terms of “MetaTopic,” based on their cell-type compositions. ( B ) Methodology used to calculate the cell type–specific score in different spot clusters. ( C ) Various downstream analyses of the SpaTopic output, including annotating spatial domains based on cell-type compositions, domain interactions, and spatial patterns of gene expression, and comparing spatial domains in terms of MetaTopics. SpaTopic can efficiently capture known spatial domains such as TBs and TLSs. ( D and E ) Performance of different methods for spatial domain calling. The evaluation is performed using 12 SRT datasets with manual annotation. ARI , adjusted Rand index.

Fig. 1 from Yuelei Zhang et al., Science Advances (2024). Licensed CC BY-NC. Source: 10.1126/sciadv.adp4942.

Science Advances (2024) · 22 citations

Read the original → https://doi.org/10.1126/sciadv.adp4942

Deep learning-based cell-specific gene regulatory networks inferred from single-cell multiome data

scMultiomeGRN is a deep learning method that fuses two single-cell measurements — RNA levels and chromatin accessibility — to work out which transcription factors regulate which target genes. Single-cell data is notoriously noisy, so the framework is built to handle missing values. The payoff is a more trustworthy map of gene regulatory networks in individual cell types.

Conceptual illustration for: Deep learning-based cell-specific gene regulatory networks inferred from single-cell multiome data

Nucleic Acids Research (2025) · 46 citations

Read the original → https://doi.org/10.1093/nar/gkaf138

Single-cell transcriptomics and functional analyses identified SPIN.DOC as a regulator for Wnt signaling and cancer stemness in colorectal cancer.

This study mines single-cell RNA sequencing data from colorectal cancer patients to look into SPIN.DOC, a gene that shows up at unusually high levels in tumor cells. The authors connect SPIN.DOC to activation of the Wnt signaling pathway and to cancer stem cell traits, pointing to it as a possible driver of tumor growth. The work gives others a foothold for follow-up functional studies and potential therapeutic targeting.

Conceptual illustration for: Single-cell transcriptomics and functional analyses identified SPIN.DOC as a regulator for Wnt signaling and cancer stemness in colorectal cancer.

Human cell (2026)

Read the original → https://doi.org/10.1007/s13577-026-01442-9

HKDC1 as a Prognostic and Tumor Microenvironment‑Associated Biomarker in Colorectal Cancer: An Integrated Multi‑omics and Spatial Transcriptomic Study.

Using multiple public datasets plus spatial maps, this study makes the case that HKDC1, a relatively recently characterized enzyme involved in glucose metabolism, is a meaningful prognostic biomarker in colorectal cancer. The researchers show where HKDC1 is active inside the tumor and how its expression tracks with the surrounding cellular neighborhood. The broader point: metabolic enzymes can do more than process fuel — they also report on tumor state.

Conceptual illustration for: HKDC1 as a Prognostic and Tumor Microenvironment‑Associated Biomarker in Colorectal Cancer: An Integrated Multi‑omics and Spatial Transcriptomic Study.

Digestive diseases and sciences (2026)

Read the original → https://doi.org/10.1007/s10620-026-10182-7

Delineating copy number and clonal substructure in human tumors from single-cell transcriptomes

CopyKAT estimates large-scale DNA copy number changes — gains and losses of chromosome pieces — straight from single-cell RNA sequencing. That lets researchers separate malignant cells from normal ones in a tumor sample and uncover subclonal structure without dedicated DNA sequencing. It has settled into a standard step in many single-cell tumor studies.

Conceptual illustration for: Delineating copy number and clonal substructure in human tumors from single-cell transcriptomes

Nature Biotechnology (2021) · 735 citations

Read the original → https://doi.org/10.1038/s41587-020-00795-2

Quick links

Paper Venue / Year Link
SPACEL: deep learning-based characterization of spatial transcriptome architectures Nature Communications (2023) Read →
SpaTopic: A statistical learning framework for exploring tumor spatial architecture from spatially resolved transcriptomic data Science Advances (2024) Read →
Deep learning-based cell-specific gene regulatory networks inferred from single-cell multiome data Nucleic Acids Research (2025) Read →
Single-cell transcriptomics and functional analyses identified SPIN.DOC as a regulator for Wnt signaling and cancer stemness in colorectal cancer. Human cell (2026) Read →
HKDC1 as a Prognostic and Tumor Microenvironment‑Associated Biomarker in Colorectal Cancer: An Integrated Multi‑omics and Spatial Transcriptomic Study. Digestive diseases and sciences (2026) Read →
Delineating copy number and clonal substructure in human tumors from single-cell transcriptomes Nature Biotechnology (2021) Read →

Spatial Transcriptomics Spatial Omics Single-Cell Transcriptomics scRNA-seq Genomics Cancer Oncology