PIASO tutorial and processed single-cell omics datasets

This record contains a collection of single-cell omics datasets in AnnData HDF5 (.h5ad), cytome (.cytome) and other formats associated with PIASO (Precise Integrative Analysis of Single-cell Omics), a Python toolkit for single-cell data analysis (https://github.com/genecell/PIASO). The record bundles two categories of files: (i) input datasets used in the official PIASO tutorials, and (ii) PIASO-processed datasets containing derived results (embeddings, cluster labels, marker genes, integrated representations) produced by running PIASO modules on those inputs. Contents. This version contains single-cell and single-nucleus RNA-seq datasets: reference and query datasets used to illustrate normalization (INFOG), dimensionality reduction and integration (GDR), marker gene identification (COSG), cell-type annotation, and gene set scoring; the RNA modality of several snMultiome experiments; and a pre-computed PIASOmarkerDB marker table. The scATAC-seq and spatial transcriptomics datasets used in the corresponding PIASO tutorials are not yet included here; the tutorials at https://piaso.org name their sources. New in this version: the cytome format Five RNA datasets are now also provided as .cytome files. A .cytome is a single SQLite-backed file that is read in chunks rather than loaded whole, so a 1.5-million-nucleus dataset can be normalized, embedded and clustered without holding it in memory, and the same file is read natively from both Python and R with no conversion step between them. Each cytome carries one measurement, RNA_counts, holding raw UMI counts that were verified as non-negative integers at conversion time. Normalized and scaled layers are deliberately left out: they are a deterministic function of the counts and the parameters, so they recompute, and storing them would give a reader two matrices to choose between when only one is authoritative. Cell and gene annotations, embeddings and graphs from the source files are preserved. Python import cytome ds = cytome.open("humanlifespan_pfc_rna.cytome") ds.n_cells, ds.n_genes for chunk, rows in ds.iter_chunks(modality="RNA", layer="counts", batch_size=4096): ... # chunk is a scipy CSR matrix R install.packages("cytome", repos = "https://genecell.r-universe.dev") sce <- cytome::read_cytome("humanlifespan_pfc_rna.cytome") All five are cytome format version 1.0.0, written by cytome 0.2.4; each file records this in its own manifest, so it can be checked without trusting this page. Readers: https://github.com/genecell/cytome (Python) and https://github.com/genecell/cytome-r (R). The file MD5SUMS_cytome_files lists checksums for the five cytome files. The cytome files are format conversions of published datasets. Please cite the original study for whichever dataset you use; each is named below. Files in this version cytome format (new in this version) sea_ad_mtg_20k.cytome (269 MB) — 20,000 nuclei × 36,601 genes. SEA-AD middle temporal gyrus, human, subsampled with the neurotypical reference donors excluded. 84 donors aged 65 to 102 at death, 11,345 nuclei from donors with no dementia and 8,655 from donors with dementia. SEA-AD taxonomy at three depths (Class 3, Subclass 24, Supertype 139), each with mapping confidence, plus the full neuropathology panel (Thal, Braak, CERAD, CAA, Lewy, LATE), APOE genotype and CASI/MMSE/MoCA scores. X_scVI and X_umap embeddings. Source: Gabitto, M.I., Travaglini, K.J., Rachleff, V.M. et al. Integrated multimodal cell atlas of Alzheimer's disease. Nat Neurosci 27, 2366–2383 (2024). https://doi.org/10.1038/s41593-024-01774-5 adult_cortex_multiome_rna.cytome (192 MB) — 17,412 nuclei × 26,205 genes, mm10. The gene-expression half of an adult mouse cortex 10x Multiome experiment. Five libraries differing in nucleus preparation, so the dataset carries a real batch axis. 20-label cortical cell type annotation (CellTypes) plus an independent label-transfer annotation (CellTypes_TACCO), several Leiden clusterings, per-nucleus QC from cellranger-arc including the ATAC-side fragment, TSS and peak counts, and X_pca, X_pca_harmony and three UMAPs. Source: Bravo Gonzalez-Blas, C., De Winter, S., Hulselmans, G. et al. SCENIC+: single-cell multiomic inference of enhancers and gene regulatory networks. Nat Methods 20, 1355–1367 (2023). https://doi.org/10.1038/s41592-023-01938-4 allen_devvis_rna.cytome (1.4 GB) — 200,061 nuclei × 32,285 genes, mouse. Developing mouse visual cortex, 35 libraries and 31 C57BL/6J donors, fourteen ages from E15.5 to P58 sampled densely through the first two postnatal weeks, seven dissection ROIs. Allen whole-mouse-brain taxonomy at four nested depths with mapping probabilities: class (15), subclass (40), cluster (142), subcluster (634). Immature-neuron, glioblast, radial glia and intermediate-progenitor populations are labelled. Source: Gao, Y., van Velthoven, C.T.J., Lee, C. et al. Continuous cell-type diversification in mouse visual cortex development. Nature 647, 127–142 (2025). https://doi.org/10.1038/s41586-025-09644-1 humandevcx_38_rna.cytome (1.1 GB) — 213,090 nuclei × 36,601 genes, human. Developing human cortex across 38 samples, spanning first trimester through adolescence in five groups (first trimester 20,362 nuclei, second 63,396, third 22,603, infancy 52,125, adolescence 54,604). Annotations at two depths: Subclass (11 broad classes) and Type (33 fine labels resolving excitatory neurons by layer and maturation state and radial glia by vRG/oRG/tRG). Source: Wang, L., Wang, C., Moriano, J.A. et al. Molecular and cellular dynamics of the developing human neocortex. Nature 647, 169–178 (2025). https://doi.org/10.1038/s41586-024-08351-7 humanlifespan_pfc_rna.cytome (25.7 GB) — 1,501,089 nuclei × 38,606 genes, human. Prefrontal cortex across the adult lifespan, 357 samples, ages 15 to 90 (median 46). Two cohorts (NABEC 942,845 nuclei, HBCC 558,244) from four brain banks, 968,377 nuclei from male and 532,712 from female donors, ancestry recorded as European, African and White Hispanic/Latino, post-mortem interval per sample. Annotations at two depths: cell_type (7 broad) and Subclass_predicted (24 cortical subclasses) with per-nucleus confidence plus raw and neighbour-smoothed versions. Per-nucleus QC and X_svd, X_gdr, X_umap and X_gdr_umap embeddings. 4,050,255,842 non-zero entries. Source: Catching, A., Weller, C.A., Hu, F. et al. Single-nucleus multiome analysis in the human prefrontal cortex identifies gene expression and cis-regulatory elements associated with aging. Cell Reports 45, 117110 (2026). https://doi.org/10.1016/j.celrep.2026.117110 MD5SUMS_cytome_files (303 B) — MD5 checksums for the five .cytome files. AnnData and 10x formats (carried over from the previous version) SEA-AD_RNA_MTG_subsample_excludeReference_20k_piaso.h5ad (1.79 GB) — 20,000-cell subsample of the Seattle Alzheimer's Disease Brain Cell Atlas (SEA-AD), Middle Temporal Gyrus snRNA-seq, human. Raw UMI counts in layers['UMIs']. Source: Gabitto, M.I., Travaglini, K.J., Rachleff, V.M. et al. Integrated multimodal cell atlas of Alzheimer's disease. Nat Neurosci 27, 2366–2383 (2024). https://doi.org/10.1038/s41593-024-01774-5 AdultCortexMultiomeRNA_integrated_anno.h5ad (2.48 GB) — 17,412-cell integrated scRNA-seq from adult mouse cortex (P57), Multiome RNA modality across 5 batches; annotated cell types in obs['CellTypes']. Source: Bravo Gonzalez-Blas, C., De Winter, S., Hulselmans, G. et al. SCENIC+: single-cell multiomic inference of enhancers and gene regulatory networks. Nat Methods 20, 1355–1367 (2023). https://doi.org/10.1038/s41592-023-01938-4 10k_Mouse_Brain_CNIK_3p_gemx_10k_Mouse_Brain_CNIK_3p_gemx_count_sample_filtered_feature_bc_matrix.h5 (65.5 MB) — 10x Genomics 10K Mouse Brain CNIK, 3' GEM-X v4 chemistry, CellRanger-filtered matrix. Source: 10x Genomics public dataset neuron_10k_v3_filtered_feature_bc_matrix.h5 (45.4 MB) — 10x Genomics E18 mouse brain neurons, 10K cells, Chromium v3, CellRanger-filtered matrix. Source: 10x Genomics public dataset SC3_v3_NextGem_DI_Nuclei_5K_SC3_v3_NextGem_DI_Nuclei_5K_count_sample_feature_bc_matrix.h5 (19.3 MB) — 10x Genomics E18 mouse brain nuclei, 5K, Chromium Next GEM v3.1, CellRanger-filtered matrix. Source: 10x Genomics public dataset 10k_Mouse_Neurons_3p_gemx_10k_Mouse_Neurons_3p_gemx_count_sample_filtered_feature_bc_matrix.h5 (64.5 MB) — 10x Genomics E18 mouse neurons, 10K, GEM-X v4, CellRanger-filtered matrix. Source: 10x Genomics public dataset PBMCMultiomeRop2023_SAN1.h5 (73.2 MB) — Human PBMC snMultiome RNA, SAN1 sample; CellRanger-filtered 10x Multiome RNA matrix. Source: De Rop, F.V., Hulselmans, G., Flerin, C. et al. Systematic benchmarking of single-cell ATAC-sequencing protocols. Nat Biotechnol 42, 916–926 (2024). https://doi.org/10.1038/s41587-023-01881-x PBMCMultiomeRop2023_SAN2.h5 (83.7 MB) — Human PBMC snMultiome RNA, SAN2 sample; CellRanger-filtered 10x Multiome RNA matrix. Source: De Rop, F.V., Hulselmans, G., Flerin, C. et al. Systematic benchmarking of single-cell ATAC-sequencing protocols. Nat Biotechnol 42, 916–926 (2024). https://doi.org/10.1038/s41587-023-01881-x PIASOmarkerDB_AllenHumanImmuneHealthAtlas_L2_251219.csv (114.6 KB) — Pre-computed marker gene database entry for the Allen Human Immune Health Atlas at L2 annotation granularity; CSV with marker genes and specificity scores. Source: Gong, Q., Sharma, M., Glass, M.C. et al. Multi-omic profiling reveals age-related immune dynamics in healthy adults. Nature 648, 696–706 (2025). https://doi.org/10.1038/s41586-025-09686-5 Intended use. (i) Reproducing the results shown in the PIASO tutorials at https://piaso.org and https://genecell.github.io/PIASO, (ii) benchmarking PIASO against other single-cell analysis toolkits, and (iii) teaching and methods development. Primary data were obtained from publicly available sources; redistributed datasets retain their original licenses where stricter than CC BY 4.0, and per-file attribution is provided above. PIASO-processed fields (embeddings, cluster labels, marker gene results) within these files are contributed under CC BY 4.0. Software. PIASO is available at https://github.com/genecell/PIASO and can be installed via pip install piaso-tools or conda install -c conda-forge -c bioconda piaso. The cytome format libraries are at https://github.com/genecell/cytome (Python) and https://github.com/genecell/cytome-r (R). Developed in the Gord Fishell Laboratory (https://fishelllab.hms.harvard.edu) at Harvard Medical School and the Broad Institute. Citation. If PIASO is useful for your research, please consider citing: Wu, S.J., Dai, M. et al. Pyramidal neurons proportionately alter cortical interneuron subtypes. Nature (2026). https://doi.org/10.1038/s41586-025-09996-8