S3RL – Separable Spatial Single-cell Transcriptome Representation Learning for Enhanced Reconstruction of Spatial Transcriptomic Landscapes
News
2025.04.10 S3RL is now available on GitHub at https://github.com/AI4Bread/S3RL. S3RL is implemented based on PyTorch Geometric (pyG) and supports efficient training and flexible batch processing for large-scale spatial transcriptomics datasets.
The model provides enhanced spatial representation learning through the use of a Graph Transformer architecture and hyperspherical prototype clustering for clear domain separation. Please refer to Tutorials 1-5 for training strategies and batch processing guidance.
Introduction
Spatial transcriptomics enables in situ mapping of gene expression, offering insights into tissue organization and cell–cell interactions. However, its utility is limited by data sparsity and technical noise for decoding complex tissue microenvironments. Here, we introduce S3RL, a representation learning framework designed to enhance the fidelity of spatial transcriptomic data. By effectively denoising sparse measurements and amplifying biologically relevant signals, S3RL enables the recovery of fine-grained spatial expression patterns and regulatory relationships that are otherwise lost. Applied across diverse human, mouse and plant tissues, S3RL not only improves spatial domain identification and multi-slice alignment (up to 120% ARI improvement) but also uncovers previously unrecognized ligand–receptor signaling and spatial gene expression gradients critical for understanding immune-tumor crosstalk and plant developmental trajectories. These results establish S3RL as a powerful tool for extracting latent biological programs from noisy spatial transcriptomic datasets, paving the way for deeper exploration of tissue biology and disease mechanisms.
Citation
Fu, Laiyi†, Penglei Wang†, Gaoyuan Xu†, Jitao Lu, Qinke Peng, Danyang Wu* and Hequan Sun*. S3RL: Separable Spatial Single-cell Transcriptome Representation Learning for Enhanced Reconstruction of Spatial Transcriptomic Landscapes. Advanced Science, 2026: e16178.