Recent sequencing techniques have greatly enhanced our understanding of human tissues at single-cell resolution. However, many studies are limited by a small number of donors, raising concerns about generalizability. Discrepancies in cell type definitions further complicate integration efforts. Consequently, integrating multiple datasets has become crucial. Although numerous single-cell RNA sequencing results are available, larger datasets challenge traditional integration methods due to high computational demands. This study employs a Sketch strategy to construct a comprehensive prostate cell atlas, integrating all relevant open-source single-cell data.

We processed single-cell RNA sequencing files using STARsolo (version 2.7.11b) with the GRCh38-2024-A reference genome. Sample integration utilized the Seurat V5 Sketch strategy to alleviate computational burdens by sampling cells from datasets and calculating leverage scores, preserving rare subpopulations. We employed scIB for benchmarking batch correction and biological conservation across methods and developed a deep learning model for label transform, facilitating accurate data mapping.

We downloaded raw data for prostate samples from the GEO database, encompassing 12 projects, 119 samples, and 720,000 cells, including hormone-sensitive, castrate-resistant, and neuroendocrine prostate cancer samples, along with normal prostate and benign prostatic hyperplasia samples. Using the Sketch strategy, we sampled 220,000 cells and constructed a core map with cell annotations via rPCA integration. Results from Sketch-rPCA were compared to those from PCA and Harmony applied to all 720,000 cells, showing strong preservation of biological conservation. We also developed a prediction model for cell labels, enabling effective information mapping to other prostate single-cell samples.

We assessed the feasibility of the Sketch-based integration strategy in large-sample single-cell projects, saving computational and time costs while yielding integration results comparable to de novo methods. We constructed the most extensive prostate single-cell atlas to date and mapped our integration results to quarry data using deep learning models, thus providing valuable tools for researchers in related fields.