3k pbmc multiome dataset  (10X Genomics)

Journal: bioRxiv

Article Title: Chromap Suite: an open-source single-binary platform for agentic multiomic RNA + ATAC profiling

doi: 10.64898/2026.06.02.729736

Figure Lengend Snippet: Standalone wall-clock (left) and peak resident memory (right) at T = 1, 4, 24 threads versus single-threaded MACS3 v3.0.3 on the public 10x 3K PBMC Multiome ATAC channel (53.97 M deduplicated fragments). Speedup ratios annotated above libMACS3 bars. All thread counts produce a 50,274-peak narrowPeak file byte-identical to the MACS3 v3.0.3 reference; treat, lambda, and ppois bedGraphs and summit calls are likewise byte-identical (see Methods, ).

Article Snippet: The source code described in this work is freely available under permissive open-source licences: The 10x Genomics 3K PBMC Multiome dataset used for benchmarking is publicly available from 10x Genomics [ ].

Techniques:

Journal: bioRxiv

Article Title: CHAMPOLLION: Robust Multi-Omics Integration via Inverse Optimal Transport Using Paired Cells

doi: 10.64898/2026.04.28.721317

Figure Lengend Snippet: (a) Schematic representation of the first benchmark experiment. A 10X Multiome dataset is split into a bridge and a test set of pseudo-unpaired cells, with the bridge containing only 3 of the 5 cell types present in the pseudo-unpaired set. (b) LTA (higher is better) and FOSCTTM (lower is better) scores for CHAMPOLLION, MIDAS, and Seurat on the setting described in (a). Bars show the median across n = 5 random initialization seeds, with error bars indicating standard deviation; dots correspond to individual runs. (c) UMAP visualizations of the pseudo-unpaired cells for each method. Cells are colored by modality (left) and by cell type (right). (d) Transport plan inferred by CHAMPOLLION for pseudo-unpaired cells. Rows and columns are ordered by cell type, and the diagonal corresponds to the true pairs between RNA and ATAC profiles. (e) Schematic representation of the second benchmark experiment. For two paired datasets (one Multiome and one CITE-seq), half of the cells are held out as the pseudo-unpaired test set, and increasing fractions of the remainder are used as the bridge to assess the impact of its size on performance. (f) LTA and FOSCTTM scores are reported on both datasets across bridge sizes as described in (e). Lines show median performance across n = 5 seeds, while shaded regions indicate the interquartile range, and points denote individual seed results.

Article Snippet: We retrieved the 10X Multiome Human PBMC (RNA + ATAC) dataset from https://www.10xgenomics.com/datasets/pbmc-from-a-healthy-donor-no-cell-sorting-10-k-1-standard-2-0-0 .

Techniques: Standard Deviation

Journal: Bioinformatics

Article Title: CIRCE: a scalable Python package to predict cis-regulatory DNA interactions from single-cell chromatin accessibility data

doi: 10.1093/bioinformatics/btag092

Figure Lengend Snippet: CIRCE workflow and performances. (a) CIRCE workflow overview. From scATAC-seq data, co-accessibility scores are defined as regularized covariance values between DNA regions, using a graphical lasso model and pairwise distance penalties. Cis-coaccessible networks (CCANs) can then be extracted with the Louvain community detection method, as modules of DNA regions with high absolute co-accessibility scores. Co-accessibility scores in a CCAN or a DNA window of interest can then be visualized with different graphical options. (b) Correlation values between the networks obtained from CIRCE, Cicero, and metacells computation on both the BMMC and the PBMC datasets. The upper triangle of the heatmap contains the Spearman correlation, while the lower triangle contains the Pearson correlation. Colour gradient illustrates the correlation values. (c) ROC curves on the recovering DNA region—promoter interactions obtained from the PC-HiC dataset. Different preprocessed inputs are evaluated for each method: CIRCE from raw single-cell counts, binarized single-cell counts, and metacell counts from LSI dimensionality reduction space on both single-cell count matrices, and Cicero on the normalized binarized count matrices and the metacell count matrices obtained from the normalized binarized counts. (d) Running-time and memory-usage of CIRCE and Cicero when running on the single-cell or computing metacells. For CIRCE, we display performances when using multithreading (20 CPUs) and single-threading.

Article Snippet: A PBMC multiome dataset was obtained from https://www.10xgenomics.com/datasets/pbmc-from-a-healthy-donor-granulocytes-removed-through-cell-sorting-10-k-1-standard-1-0-0 .

Techniques: Single Cell