Ó³»­´«Ã½

Single-cell spatial transcriptomics can provide subcellular resolution for a deep understanding of molecular mechanisms. However, accurate segmentation and annotation remain a major challenge that limits downstream analysis. Current machine learning methods heavily rely on nuclei or cell body staining, resulting in the significant loss of both transcriptome depth and the limited ability to learn spatial colocalization patterns. Here, we propose Bering, a graph deep learning model that leverages transcript colocalization relationships for joint noise-aware cell segmentation and molecular annotation in 2D and 3D spatial transcriptomics data. To evaluate performance, we benchmark Bering with state-of-the-art methods and observe better cell segmentation accuracies and more detected transcripts across technologies and tissues. To streamline segmentation processes, we construct expansive pre-trained models, which yield high segmentation accuracy in new data through transfer learning and self-distillation. These improved capabilities enable Bering to enhance cell annotations for the rapidly expanding field of spatial omics.