PPubMed6 Oct 2025

PRTS: Predicting Single-Cell Spatial Transcriptomic Maps from Histological Images

Jingyi Wen, Lingxuan Zou, Jiying Liu, Xi Guo, Changkun Liu, Bensu Wang, Tongtong Deng, Chang Liu, Risheng Tang, Yanbin Yang, Yucheng Huang, Lijia Yang, Hui Wang, Zihao Li, Shengming Lin, Shipping Liu, Yuhu Zhang, Zhifeng Hao, Haiyu Zhou, Han Huang, Fei Ling

Abstract

High-resolution spatial transcriptomics (ST) data provide valuable insights into the molecular dynamics underlying complex biological processes. However, their widespread application remains limited due to high costs and technical challenges. Here, we present PRTS (Pathology-driven Reconstruction of Transcriptomic States), a novel framework that predicts single-cell-resolution ST data directly from histological images. Our results demonstrated that PRTS generated transcriptomic profiles for about 60,000 analyzable cell tiles per tissue section, representing an approximately 27-fold increase in analytical units compared to conventional ST spots and remarkably enhancing spatial resolution. Notably, PRTS achieves accurate cell-level transcriptomic predictions using only hematoxylin-and-eosin-stained tissue images. This method transforms costly ST technologies into a practical and scalable tool, offering a cost-efficient solution for comprehensive ST profiling in hematoxylin-and-eosin-based disease research.

A figure from PRTS: Predicting Single-Cell Spatial Transcriptomic Maps from Histological Images — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

PRTS predicts single‑cell spatial transcriptomic maps from histology images, boosting resolution ~27‑fold with modest accuracy.

What’s new

i.Introduced the PRTS framework that predicts single‑cell gene expression from H&E images.
ii.Achieved an approximate 27‑fold increase in spatial resolution over spot‑based ST methods.
iii.Demonstrated accurate gene‑expression and cell‑type predictions across mouse brain and human cancer tissues.
iv.Provided an open‑source dual‑output neural network model with both local and global histology features.

How they did it

The authors trained a dual‑output neural network that combines local (C1=192) and global (C2=384) histology features extracted from image patches to predict gene expression and binary expression status. Training used 10 epochs on ~4 million parameters, with class‑weighted loss to address gene‑expression sparsity. The workflow processes a full dataset from image input to expression matrix in about 100 minutes.

What they found

Reconstructed transcriptomes for 62,711 and 60,259 cells from 2,298 and 2,234 spots respectively (~27‑fold resolution gain).
Overall classification accuracy of 53.9% at broad cell‑type categories.
Model contains ~4.0 million parameters and completes prediction for a dataset in ~100 minutes.
Identified 19 and 17 cell subtypes in the two mouse brain datasets, and captured gene‑expression patterns consistent with ground‑truth Visium HD data.

Caveats

✕Suboptimal performance in identifying astrocytes due to their complex morphology.
✕Morphological variation from different slice preparation techniques can reduce prediction accuracy.
✕Functional annotation of tumor subpopulations remains challenging without additional cancer training data.
✕Direct one‑to‑one correspondence between predicted and manual subpopulations is hindered by differing clustering strategies.

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

Similar papers in all four archives

The same idea searched in arXiv, PubMed, bioRxiv and chemRxiv at once, so a method can lead you to where it is being applied.

AarXiv4 found

PPubMed4 found

BbioRxiv4 found

CchemRxiv3 found

Searching the other archives…