These two tracks are part of the Long-read Transcripts collection. They show the transcript isoforms of the Tabula Sapiens single-cell long-read atlas (Mantri et al. 2026), which covers 26 tissues from 12 donors, about 200,000 cells and 144 cell types. The atlas contains 854,000 isoforms. More than 330,000 of them have splice junctions that are not in GENCODE, either new combinations of known splice sites or new splice sites.
Each isoform is drawn as a transcript, with the thick part showing the open reading frame predicted by SQANTI3. Isoforms are labeled with the gene name assigned by pigeon, and the isoform ID (e.g. PB.100940.3354) can be shown as a second label on the track settings page. Both the gene name and the isoform ID can be typed into the position box. The color shows the SQANTI3 structural category, which compares the splice junctions of the isoform to the GENCODE transcripts:
| Full splice match (FSM): all splice junctions match a reference transcript | |
| Incomplete splice match (ISM): the junctions match a consecutive part of a reference transcript, but there are fewer of them | |
| Novel in catalog (NIC): all donor and acceptor sites are known, but in a new combination | |
| Novel not in catalog (NNC): at least one donor or acceptor site is not in the reference | |
| Fusion: the isoform spans two or more reference genes | |
| Genic genomic: overlaps exons and introns of a reference gene | |
| Genic intron: entirely within an intron of a reference gene | |
| Antisense: on the opposite strand of a reference gene | |
| Intergenic: no overlap with a reference gene |
The isoforms can be filtered on the track settings page by structural category, by the SQANTI3 filter result (isoform or likely artifact, see Methods), by coding potential, by predicted nonsense-mediated decay and by the number of molecules, cells or tissues in which the isoform was found. No filter is active by default, so the 162,000 isoforms that the SQANTI3 rules filter classified as likely artifacts are shown too. When more than 200 kbp are shown, the track switches to a coverage graph.
Expression is counted in molecules (UMIs), so reads that came from the same cDNA molecule are counted once. In the bar chart track, the height of each bar is the number of molecules of the isoform in a tissue per million molecules of all isoforms in that tissue.
Barcoded full-length cDNA from the 10x Genomics 3' single-cell RNA-seq reactions of the Tabula Sapiens project was used for 60 long-read libraries from 26 tissues of 12 donors, so the long reads come from the same dissociated cells as the short-read Tabula Sapiens data. After depletion of TSO artifacts, the cDNA was concatenated into 16-segment molecules with the PacBio Kinnex single-cell RNA kit and sequenced on PacBio Revio to obtain HiFi reads. Reads were split with skera, primers removed with lima, cell barcodes and UMIs extracted, corrected and deduplicated with isoseq tag, refine, correct and groupdedup, aligned to the GRCh38.p13 primary assembly with pbmm2 and collapsed into transcript models with isoseq collapse. To give each structure a single identifier across the atlas, the collapsed models of all samples were pooled, re-aligned and collapsed again. The models were classified against GENCODE v41 with pigeon and with SQANTI3 (v6.0.1, with ORF prediction, refTSS CAGE peaks and polyA motifs), and SQANTI3's rules filter labeled each model as isoform or artifact. Cells with more than 1,000 UMIs were kept, and isoforms were required to be found in at least 5 cells with at least 10 UMIs in total. Cell types were assigned with popV and the Cell Ontology. For more details, see Mantri et al. 2026.
The data files were downloaded from Figshare. The exon structures are taken from the file all_samples_pbids_isoform_structures.bed.gz, and the annotation fields from the SQANTI3 and pigeon tables (all_samples_pbids_feature_metadata_sqanti3.csv, all_samples_pbids_feature_metadata_pigeon.csv). The SQANTI3 annotation of an isoform describes the structure of the isoform in one of the samples, shown as "Annotated in" on the details page, and its ends can differ from the structure after collapsing all samples, which is the one we show. For this reason, the open reading frame predicted by SQANTI3 is drawn only when both of its ends fall into exons of the collapsed structure. This is the case for 509,719 of the 594,361 isoforms that SQANTI3 classified as coding; the remaining 84,642 are drawn without a thick part, but their ORF annotation is still shown on the details page. The UMIs and cells per tissue were summed from the per-cell isoform count matrix (all_samples_pacbio_recollapsed_raw_counts_pbids_bc_anndata_preprocessed.h5ad, 203,311 cells, 638,489,044 UMIs), and the bar chart values are these counts divided by the total number of UMIs of each tissue, times one million. All 854,410 isoforms of the input file are shown in both tracks. The commands are documented in our makeDoc file and the scripts are in the lrTrans scripts directory.
The data can be explored interactively in table format with the Table Browser or the Data Integrator and exported from there to spreadsheet or tab-separated tables. From scripts, the data can be accessed through our API, track=lrTransTabSap for the isoforms and track=lrTransTabSapExpr for the bar charts.
For automated download and analysis, the annotations are stored in bigBed files that can be downloaded from our download server. The files for these tracks are called lrTransTabSap.bb and lrTransTabSapExpr.bb. Individual regions or the whole genome annotation can be obtained using our tool bigBedToBed, which can be compiled from the source code or downloaded as a precompiled binary for your system. Instructions for downloading source code and binaries can be found here. The tool can also be used to obtain features within a given range, e.g. bigBedToBed http://hgdownload.soe.ucsc.edu/gbdb/hg38/lrTransTabSap/lrTransTabSap.bb -chrom=chr21 -start=0 -end=100000000 stdout
The original data, including the per-cell count matrices at gene, transcript and isoform level, can be downloaded from Figshare. The analysis code of the study is on GitHub.
Mantri M, Detweiler AM, Lee J, Kwon JH, Zhou A, Tong L, Jones RC, Neff NF, Tabula Sapiens Consortium, Quake SR. Single-cell splice isoform usage reveals distinct axes of cellular identity and senescence. bioRxiv. 2026 Sep 14.