A collection of AI-ready single-cell transcriptomic datasets for single-cell foundation models

Single-cell foundation models (scFMs) trained on large-scale single-cell omics data, especially single-cell transcriptomic datasets, are revolutionizing our understanding of cellular heterogeneity and biological systems. However, the lack of large-scale, standardized, and artificial intelligence-ready (AI) single-cell data with consistent cell-type annotations limits model reproducibility and generalization. Here, we introduce PreDigs-AI, a high-quality, multi-granular single-cell collection tailored for scFM training and benchmarking. Datasets are uniformly processed and stored in standardized h5ad format for immediate use with models like scGPT and Geneformer.