STRESS: Spatial Transcriptome Resolution Enhancing Method based on the State Space Model

The widespread application of spatial resolved transcriptomics (SRT) has provided a wealth of data for characterizing gene expression patterns within the spatial microenvironments of various tissues. However, the inherent resolution limitations of SRT in most published data restrict deeper biological insights. To overcome this resolution limitation, we propose STRESS, the first deep learning method designed specifically for resolution enhancement tasks using only SRT data. By constructing a 3D structure that integrates spatial location information and gene expression levels, STRESS identifies interaction relationships between different locations and genes, predicts the gene expression profiles in the gaps surrounding each spot, and achieves resolution enhancement based on these predictions. STRESS has been trained and tested on datasets from multiple platforms and tissue types, and its utility in downstream analyses has been validated using independent datasets. We demonstrate that this resolution enhancement facilitates the identification and delineation of spatial domains. STRESS holds significant potential for advancing data mining in spatial transcriptomics.空间分辨转录组学（Spatial Resolved Transcriptomics, SRT）的广泛应用，为刻画多种组织空间微环境内的基因表达模式提供了海量数据。然而，多数已发表数据中SRT固有的分辨率局限，制约了更深入的生物学解析。为突破这一分辨率瓶颈，本文提出STRESS——首个专为仅依托SRT数据开展分辨率增强任务设计的深度学习方法。STRESS通过构建整合空间位置信息与基因表达水平的三维结构，识别不同空间位点与基因间的相互作用关系，预测每个斑点周围间隙的基因表达谱，并基于上述预测实现分辨率增强。本方法已在多平台、多组织类型的数据集上完成训练与测试，并通过独立数据集验证了其在下游分析中的应用价值。研究表明，该分辨率增强操作有助于识别并勾勒空间结构域。STRESS在推动空间转录组学数据挖掘领域展现出巨大应用潜力。