Genes interact with each other in complex regulatory programs to determine cellular func tions. A gene regulatory network (GRN) provides a systems-level structure of these in teractions, enabling the integrated analysis of gene regulations. Accurate GRN inference is therefore essential for uncovering biological mechanisms and for investigating disease pathogenesis and potential therapeutic strategies. Recent advances in single-cell sequencing enabled GRN inference at single-cell resolution, but such data are inherently sparse and noisy, yielding low signal-to-noise ratios that com plicate reliable reconstruction of GRN. In addition, most existing methods rely on prior GRNassembled from mixed biological contexts, which may include interactions irrelevant to a given cell type, or miss true condition-specific interactions. Accurately inferring GRNs that reflect each biological condition is thus a critical challenge. To address this, we introduce INSCAPE (INference of condition-SpeCific gene regulatory networks through subgrAph toPology lEarning), an end-to-end framework that integrates single-cell data with prior GRN knowledge to infer condition-specific regulatory relation ships. INSCAPEincludes built-in preprocessing of single-cell data that supports multi-ome integration when available. It introduces a subgraph-based topological regularization strat egy, which enables the model to effectively overcome the incomplete and noisy nature of prior GRNs. In addition, INSCAPE employs a multi-view attention mechanism that cap tures regulatory signals from both feature-based and topology-based perspectives, enhanc ing the discovery of novel regulatory interactions. We evaluated INSCAPE on six single-cell datasets?four scRNA-seq and two multi-ome spanning five cell types and one disease condition. Across all datasets, INSCAPE consis tently achieved superior performance compared to existing baseline methods.