SpaGene: A Deep Adversarial Framework for Spatial Gene Imputation

Integrating transcriptome-wide single-cell gene expression data with spatial context substantially enhances our understanding of tissue biology, cellular interactions, and disease progression. Although single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression data, it lacks crucial spatial context, whereas spatial transcriptomics techniques offer spatial resolution but are limited in the transcriptomics coverage. To address these limitations, integrating scRNA-seq and spatial transcriptomics data is essential. We introduce SpaGene, a deep learning framework designed to integrate scRNA-seq data and spatial transcriptomics data. SpaGene consists of 2 encoder-decoder pairs combined with 2 translators and 2 discriminators to effectively impute missing gene expression within spatial transcriptomics datasets. We benchmarked SpaGene against 6 representative methods across diverse datasets. Across dataset pairs under a controlled gene-holdout evaluation protocol, SpaGene improves average Pearson correlation coefficient and cell-wise structural similarity index and reduces root mean squared error compared to the evaluated baselines, indicating more accurate recovery of held-out spatial gene expression. Application of our model to lung tumor tissue revealed spatial patterns consistent with immune cell enrichment at tumor boundaries, restricted myeloid cell presence in adjacent normal regions, and improved detection sensitivity of microenvironment-driven pathways linked to immune neighborhoods. These findings provide insight into immune exclusion and tumor-immune interactions motivating further biological validation.