PH2ST:ST-Prompt Guided Histological Hypergraph Learning for Spatial Gene Expression Prediction

TL;DR

PH2ST tackles the challenge of cost and coverage in spatial transcriptomics by introducing an inference-time, prompt-guided framework that predicts spatial gene expression from H&E images using limited ST signals. It advances a dual-scale hypergraph encoder to model local and global tissue context and employs cross-attention to align histology with ST prompts, enabling robust predictions across diverse samples. The approach achieves state-of-the-art results on two public ST datasets, with extensive ablations and analyses under realistic prompt sampling strategies, and demonstrates practical potential for ST imputation, super-resolution, and local-to-global mapping. By enabling accurate, low-cost spatial gene expression mapping, PH2ST has meaningful implications for scalable clinical and research workflows in biomedical contexts.

Abstract

Spatial Transcriptomics (ST) reveals the spatial distribution of gene expression in tissues, offering critical insights into biological processes and disease mechanisms. However, the high cost, limited coverage, and technical complexity of current ST technologies restrict their widespread use in clinical and research settings, making obtaining high-resolution transcriptomic profiles across large tissue areas challenging. Predicting ST from H\&E-stained histology images has emerged as a promising alternative to address these limitations but remains challenging due to the heterogeneous relationship between histomorphology and gene expression, which is affected by substantial variability across patients and tissue sections. In response, we propose PH2ST, a prompt-guided hypergraph learning framework, which leverages limited ST signals to guide multi-scale histological representation learning for accurate and robust spatial gene expression prediction. Extensive evaluations on two public ST datasets and multiple prompt sampling strategies simulating real-world scenarios demonstrate that PH2ST not only outperforms existing state-of-the-art methods, but also shows strong potential for practical applications such as imputing missing spots, ST super-resolution, and local-to-global prediction, highlighting its value for scalable and cost-effective spatial gene expression mapping in biomedical contexts.

Figures (6)

• Figure 1: Illustration of (a) traditional ST prediction and (b) the proposed inference-time prompting paradigm.
• Figure 2: Overview of PH2ST. (a) UNI extracts histological features at two spatial scales. (b) Dual-scale hypergraphs capture high-order spatial relationships, with global fusion enabled by Transformer blocks. (c) A learnable projection module encodes the spatial gene expression matrix into ST-prompt embeddings. Four prompt sampling strategies used during inference are illustrated. (d) A cross-attention mechanism fuses the ST-prompt with neighboring features. The combined representations are used for final prediction.
• Figure 3: Comparison of PCC on HER2+ and cSCC datasets using non-parametric test.
• Figure 4: Visualization of WSI thumbnails, ST ground truths, and model predictions on two datasets. A representative sample was randomly selected from each dataset. (a) Marker genes: $\textit{TMSB10}$, $\textit{CISD3}$, $\textit{CD74}$ and $\textit{COL6A2}$ on HER2+ dataset. (b) Marker genes: $\textit{MT-CO2}$, $\textit{PKP1}$, $\textit{RPS5}$ and $\textit{SFN}$ on cSCC dataset.
• Figure 5: Ablation results (PCC and CCC) on the cSCC dataset. (a) Effect of fine-tuning on baseline methods and contribution of the proposed dual-scale histological hypergraphs. (b–c) Influence of ST-prompt proportions during training and testing on model performance.