Model comparison on different cancers.

MicroRNAs (miRNAs) are critical regulators of gene expression in cancer biology, yet their spatial dynamics within tumor microenvironments (TMEs) remain underexplored due to technical limitations in current spatial transcriptomics (ST) technologies. To address this gap, we present STmiR, a novel XGBoost-based framework for spatially resolved miRNA activity prediction. STmiR integrates bulk RNA-seq data (TCGA and CCLE) with spatial transcriptomics profiles to model nonlinear miRNA-mRNA interactions, achieving high predictive accuracy (Spearman’s ρ > 0.8) across four major cancer types (breast, lung, ovarian, prostate), with performance further confirmed through direct comparison with experimentally measured miRNA expression in an independent spatial transcriptomics dataset. Applied to 10X Visium ST datasets from nine cancers, STmiR identifies six pan-cancer conserved miRNAs (e.g., hsa-miR-21, hsa-let-7a) consistently ranked in the top 40 across malignancies, and uncovers cell-type-specific regulatory networks in fibroblasts, B cells, and malignant cells. A breast cancer case study demonstrates STmiR’s utility in uncovering biologically relevant miRNA-target relationships and their association with key cancer pathways. By enabling spatial mapping of miRNA activity, STmiR provides a transformative tool to dissect miRNA-mediated regulatory mechanisms in cancer progression and TME remodeling, with implications for biomarker discovery and precision oncology.