Fig. 4

txci-ATAC-seq generates high-quality single-cell ATAC-seq data on multiple tissues in parallel at scale. a Well assignment showing the multiplexing of primary samples. Rows 7 and 8 provide an estimate of the empirical collision rate for each sample by mixing human lung nuclei with mouse nuclei isolated from each corresponding tissue. b-d The comparison of quality metrics between sciATAC-seq and txci-ATAC-seq for each cell in mouse lung and liver tissue. The (b) FRiDHS, (c) TSS enrichment score, and (d) estimated complexity (on a log₁₀ scale) indicate the performance of single-cell ATAC-seq methods. The numbers over the violin plots reflect the fold-change in median compared to txci-ATAC-seq. e UMAP visualization of mouse lung nuclei (n = 73,280) integrating two replicates across two loading inputs. Nuclei are colored by their predicted cell type. f UMAP visualization of mouse liver nuclei (n = 63,429) integrating two replicates across two loading inputs. Abbreviations: AM, alveolar macrophages; AT1, alveolar type 1 epithelial cells; AT2, alveolar type 2 epithelial cells; avlEC, arterial/venous/lymphatic endothelial cells; B/T sub, B and T cell subpopulation; cEC, capillary endothelial cells; Col13 + FB, collagen type XIII α 1 chain positive fibroblasts; Col14 + FB, collagen type XIV α 1 chain positive fibroblasts; DC/IM/cMono, dendritic cells/interstitial macrophages/classical monocytes; EC, endothelial cells; GB, germinal B cells; Hep, hepatocytes; HPC/Cho, hepatic progenitor cells/cholangiocytes; KC/Mono, Kupffer cells/monocytes; lsEC, liver sinusoidal endothelial cells; Lym, lymphocytes; Mes, mesothelial cells; MyoFB, myofibroblasts; ncMono, nonclassical monocytes; Peri, pericytes; SMC, smooth muscle cells, vEC, venous endothelial cells, vSMC, vascular smooth muscle cells. g Theoretical and empirical number of deconvolutable cells recovered across various nuclei loading inputs for molecular and cellular hashing strategies. The simulated cell recovery for either molecular hashing at different numbers of Tn5 barcodes (black and gray lines) or cellular hashing (blue line) strategies were compared with the observed cell recovery obtained from the txci-ATAC (indexed with 96 Tn5 barcodes, red dots), dsciATAC (indexed with 48 Tn5 barcodes, slate blue dots), and SNuBar (blue dots) datasets. For txci-ATAC and dsciATAC, the data were processed using the same pipeline, and the cell threshold was determined using K-means clustering. For SNuBar, the recovery values were taken directly from the original paper. The lower panel zooms in on the range highlighted by the dashed box in the upper panel. h Unusable-to-usable cell ratio derived from the simulated data presented in panel (g). Unusable refers to cells that had to be discarded because they were in unresolvable multiplets. Usable cells include singlets and cells that can be unambiguously demultiplexed from multiplets for molecular hashing strategies. The colors are consistent with panel (g). i Estimated collision rate calculated from the cells determined by K-means clustering for both txci-ATAC and dsciATAC datasets across various loading inputs. A mixture of human K562 and murine 3T3 cells was used to identify multiplets in the dsciATAC datasets (slate blue). For txci-ATAC, the collision rate was evaluated using either a mixture of mouse and human lung cells (red) or a mixture of mouse liver and human lung cells (orange). The data points from the txci-ATAC datasets were jittered to enhance the visibility of individual values