ST-DAI: Single-shot 2.5D Spatial Transcriptomics with Intra-Sample Domain Adaptive Imputation for Cost-efficient 3D Reconstruction.

Three-dimensional spatial transcriptomics (ST) is constrained by the high cost of fully profiling every serial tissue section. Existing histology-to-expression approaches can reduce within-sample measurement, but they generally require large external training cohorts and may be affected by cross-specimen domain shift. We introduce ST-DAI, a sample-specific framework that couples a cost-efficient 2.5D acquisition protocol with intra-sample domain-adaptive imputation. One reference section is fully measured, while adjacent target sections are sparsely measured to provide specimen-specific molecular anchors. ST-DAI first aligns the reference and target sections, pretrains a sparse-to-dense Pseudo Map Network on the reference section, and then performs Fast Multi-Domain Refinement using Parameter-Efficient Domain-Alignment Layers and confidence-weighted pseudo-supervision. A Data Consistency Operation preserves all experimentally observed target values in the final reconstruction. Across six Xenium breast-cancer and six Human DLPFC 10x Visium section settings, ST-DAI achieves the best reconstruction metrics among the evaluated methods under their respective acquisition protocols. Additional analyses further show improved preservation of morphology-derived five-class nuclear composition, differential-expression contrasts, marker localization, and spatial-domain organization. These results support sample-specific 2.5D acquisition as a practical strategy for reducing the measurement burden of volumetric ST while retaining quantitative and biologically interpretable spatial expression structure.