Journal: bioRxiv

Article Title: PRISM: Niche-informed Deciphering of Incomplete Spatial Multi-Omics Data

doi: 10.64898/2026.02.03.703456

Figure Lengend Snippet: a , Experimental schematic for generating spatial multi-omics data from serial human lymphoid organ (tonsil) sections, with transcriptomics profiling (Step 1) and spatial protein co-profiling (Step 2). b , Analysis workflow: incomplete spatial multi-omics inputs (containing an unregistered proteomics region) are processed by PRISM to identify spatial domains and impute protein signals in unregistered locations. c , Simulation of FOV-induced incompleteness by masking a contiguous horizontal region (50%) of the proteomics modality while retaining full transcriptomics coverage (the masked area is treated as unregistered). d , Ground-truth spatial domain annotation of the tonsil section (connective & epithelial tissue, germinal center, lymphoid follicle, and tonsillar parenchyma). e , Comparison of spatial domain identification results between PRISM and advanced spatial multi-omics baseline method. f , Quantitative comparison of domain identification using AMI, V-measure, homogeneity, NMI and ARI. g , Spatial visualization of representative proteins: ground truth measurements compared with predictions from non-spatial translators and spatial methods. h , Overall protein imputation performance is summarized by PCC and SPCC across methods. i , Compare the protein-related distributions (PCC and SPCC) of different methods using box plots. Each dot in the box represents one protein. The statistical significance of PRISM and other methods were indicated using * (*: p-value<0.05, **: p-value<0.01, ***: p-value<0.001, ****: p-value<0.0001). j , Robustness analysis under varying FOV overlap rates (from 90% down to 10%), illustrating performance trends as the registered region decreases. k , Robustness analysis regarding the spatial position of the registered window: a sliding-window scheme with fixed overlap (50%) evaluated across multiple positions.

Article Snippet: To profile these tissues in situ across molecular layers, the Spatial Transcriptomics-Proteomics (STP) workflow is designed to simultaneously capture spatially resolved transcript and protein expression, while preserving the specificity and sensitivity of both modalities.

Techniques: Biomarker Discovery, Transcriptomics, Comparison

Journal: bioRxiv

Article Title: PRISM: Niche-informed Deciphering of Incomplete Spatial Multi-Omics Data

doi: 10.64898/2026.02.03.703456

Figure Lengend Snippet: a , Workflow schematic: Visium RNA (55 µm spots) and MALDI-MSI (100 µm pixels) generated via incompatible platforms are co-registered using MAGPIE, yielding an incomplete spatial multi-omics dataset with resolution-induced gaps. PRISM then infers spatial domains and imputes MSI signals in unregistered locations. b , H&E image of the human striatum section with three distinct coronal regions (Cd1-Cd3) labeled. c , Illustration of cross-modal correspondence after registration, showing RNA spots, MSI pixels and matched pairs within the aligned coordinate system. d , Ground-truth spatial domain annotations for the three regions. NA: non-annotated or filtered areas (including mounting artifacts). e , Visual comparison of spatial domains identified by PRISM versus state-of-the-art baselines under the resolution-induced incompleteness. f , Quantitative benchmarking of domain identification performance across sections using five established metrics. g , Spatial map highlighting the mounting-artifact region and its assignment relative to other clusters. h , Differential gene expression summary for the mounting-artifact region versus inferred clusters, showing representative marker genes with dot size indicating the fraction of cells and color indicating mean expression.

Article Snippet: To profile these tissues in situ across molecular layers, the Spatial Transcriptomics-Proteomics (STP) workflow is designed to simultaneously capture spatially resolved transcript and protein expression, while preserving the specificity and sensitivity of both modalities.

Techniques: Generated, Biomarker Discovery, Labeling, Comparison, Gene Expression, Marker, Expressing