Project overview
This report is a sample deliverable built on public data to show how HST GENOMICS delivers a standard single-cell RNA-seq analysis. Data: the 10x Genomics public dataset "PBMC 10k v3" (healthy donor peripheral blood mononuclear cells, Chromium v3 chemistry), starting from the Cell Ranger filtered feature-barcode matrix.
11,037
cells after QC
22
clusters
1,950
median genes / cell
9.0%
median mito %
Methods
- Software: scanpy 1.12.4, anndata 0.13.4, Leiden via igraph 1.0.0.
- QC: cells with ≥ 200 genes, < 6,000 genes and < 20% mitochondrial reads kept; genes detected in ≥ 3 cells kept. 11,769 cells in, 11,037 retained.
- Normalisation: counts per cell scaled to 1e4 then log1p; 2,000 highly variable genes (seurat_v3); scaling and PCA (50 components).
- Clustering and embedding: kNN graph (k=15) on 30 PCs, Leiden clustering (resolution 0.8), UMAP.
- Markers: Wilcoxon rank-sum test of each cluster versus the rest. Annotation: the top 30 significantly up-regulated markers of each cluster (log2FC > 0.5, adjusted P < 0.05) were matched against 11 canonical marker sets; a type is assigned only with at least two matching genes. Clusters matching two of the T/NK, B and myeloid lineages are labelled doublet-like. A cluster is never labelled CD4 T or CD8 T when CD4 or CD8A is significantly down-regulated in it. Clusters with no matching marker among their top 30 are named only if a marker-set mean score is at least 0.5 (shown as score in the Basis column of Table 1); otherwise they stay Unassigned and keep their number. Doublets: scored with scrublet, flagged but not removed.
Quality control
After filtering: median 1,950 genes and 6,729 UMIs per cell; median mitochondrial fraction 9.0%.
Results
| Cluster | Annotation | Cells | Predicted doublets | Basis | Top 5 markers |
|---|---|---|---|---|---|
| 0 | CD4 T | 1620 | 1% | CD3D,CD3E,IL7R | IL7R, LDHB, TRAC, TPT1, RPS29 |
| 1 | CD8 T | 684 | 3% | CD3D,CD3E | KLRB1, GZMK, KLRG1, NKG7, GZMA |
| 2 | HSPC | 38 | 0% | CD34,PRSS57,SOX4 | RPS24, RPL7A, HNRNPA1, SOX4, ITM2C |
| 3 | CD4 T | 1118 | 0% | score 0.77 | RPL30, RPL32, RPS3A, RPS14, RPS12 |
| 4 | DC | 193 | 11% | FCER1A,CST3,CLEC10A | HLA-DPA1, CST3, HLA-DPB1, HLA-DRA, HLA-DRB1 |
| 5 | CD14 Mono | 1270 | 0% | score 2.65 | LGALS2, CPVL, PSAP, NEAT1, FGL2 |
| 6 | CD14 Mono | 1671 | 1% | CD14,LYZ,S100A8,S100A9 | S100A8, S100A12, S100A9, MNDA, VCAN |
| 7 | NK | 608 | 4% | GNLY,NKG7,KLRD1 | GNLY, NKG7, PRF1, KLRD1, CTSW |
| 8 | B | 987 | 0% | MS4A1,CD79A,CD79B | IGHM, IGHD, CD79A, CD37, TCL1A |
| 9 | B | 481 | 5% | MS4A1,CD79A,CD79B | MS4A1, BANK1, CD79A, CD37, CD74 |
| 10 | CD8 T | 595 | 2% | CD3D,CD3E,CD8A,CD8B | CCL5, NKG7, IL32, CTSW, GZMM |
| 11 | CD8 T | 372 | 1% | CD8A,CD8B | CD8B, RPS12, RPS3A, RPL32, RPS6 |
| 12 | pDC | 83 | 16% | LILRA4,IL3RA | UGCG, CCDC50, PLD4, IRF8, TCF4 |
| 13 | CD16 Mono | 394 | 11% | FCGR3A,MS4A7,LST1 | LST1, FCGR3A, AIF1, SMIM25, COTL1 |
| 14 | Platelet | 179 | 5% | PPBP,PF4,GP9 | PF4, PPBP, NRGN, CAVIN2, TUBB1 |
| 15 | CD14 Mono | 138 | 1% | CD14,LYZ,S100A8,S100A9 | TYMP, MX1, TNFSF13B, MX2, IFI44 |
| 16 | Plasma | 19 | 5% | MZB1,JCHAIN,TNFRSF17 | MZB1, SUB1, JCHAIN, SSR4, PPIB |
| 17 | B | 95 | 99% | MS4A1,CD79A,CD79B | MS4A1, IGHM, CD79A, IGHD, LINC00926 |
| 18 | CD4 T | 202 | 0% | CD3D,CD3E | IL32, TRAC, B2M, HLA-A, CD3D |
| 19 | Doublet-like (B/CD14 Mono) | 69 | 97% | B: MS4A1,CD79A,CD79B; CD14 Mono: CD14,S100A8,S100A9 | MS4A1, CD79A, IGHM, RALGPS2, CD79B |
| 20 | CD14 Mono | 205 | 94% | CD14,LYZ,S100A8,S100A9 | S100A12, S100A9, S100A8, CD14, VCAN |
| 21 | DC | 16 | 0% | CST3,CLEC9A | CLEC9A, IDO1, C1orf54, HLA-DPA1, HLA-DPB1 |
Table 1. Cluster annotation and markers (full table: tables/cluster_markers_wilcoxon.csv).
Doublets: scrublet flags 5.0% of cells. The following clusters have more than 50% predicted doublets but markers of a single cell type, so their labels are kept and counted; in a real project they would be reviewed against sample information: cluster 17 (B, 99%), cluster 20 (CD14 Mono, 94%).
Deliverables
- tables/cell_metadata_umap.csv: per-cell QC metrics, cluster, cell type and UMAP coordinates
- tables/cluster_markers_wilcoxon.csv: full marker table per cluster
- tables/cell_type_proportions.csv, tables/cluster_marker_scores.csv: proportions and annotation scores
- run_scrna.py, report_lib.py, requirements.txt, README.txt: complete re-runnable code, parameters and instructions
- figures/*.png: publication figures; processed h5ad object available on request
- run_scrna.py: complete, re-runnable analysis code with parameters
Methods paragraph (for the manuscript)
The Cell Ranger filtered matrix was processed with scanpy 1.12.4. Cells with ≥ 200 and < 6,000 detected genes and < 20% mitochondrial reads, and genes detected in ≥ 3 cells, were retained. Counts were normalised to 10,000 per cell and log1p-transformed; 2,000 highly variable genes were selected (seurat_v3) and scaled before PCA. A k-nearest-neighbour graph (k = 15) on the first 30 principal components was clustered with the Leiden algorithm (resolution 0.8) and visualised with UMAP. Cluster markers were identified with the Wilcoxon rank-sum test; cell types were assigned when at least two canonical marker genes were among the top 30 markers of a cluster, and clusters with conflicting lineage markers were labelled doublet-like. Doublets were scored with scrublet.