nsclc

nsclc — Homo sapiens · lung · scRNA-seq.

• Cells: 6122 · Genes: 20213 · Clusters: 14 (seurat_clusters)
• Source: tidx package (20213 genes, cluster col seurat_clusters)
• Embedding: UMAP

This report covers a single-cell RNA-seq dataset from human lung. It has 6,122 cells, and 20,213 genes were measured in each cell. The cells were grouped into 14 clusters by the seurat_clusters column and are shown on a UMAP embedding. The dataset does not record a condition, genotype or sample column. The report therefore describes one population of cells and does not compare groups.

The clusters vary widely in size. Cluster 0 alone holds 2,271 cells (37.1% of the total). Clusters 0 and 1 together hold 3,239 cells (52.9%). The next nine clusters range from 601 cells down to 142. The three smallest clusters are cluster 11 (69 cells), cluster 12 (43 cells) and cluster 13 (43 cells), which together hold 155 cells (2.5%). The largest cluster is about 53 times the size of the smallest. Per-cluster statistics for clusters 12 and 13 rest on 43 cells each and should be read with that in mind. The facts record no PCA components (pca_components is empty), so the trajectory section is not based on a stored PCA.

A human has reviewed all 14 cluster labels. The cell-type annotation section, which is also reviewed, is the authority on what each cluster is. The six sections after it are still AI drafts: the Gene Ontology and pathway enrichment, the marker heatmap, the QC summary, the trajectory and the UMAP overview. Their numbers are computed from the data, but their wording has not yet been checked by a person. The table below lists each section and its current status.

• Cells: 6122 · Genes (matrix rows): 20213 · Matrix density: 0.0841
• Clusters (seurat_clusters): 14 — sizes 43–2271

Cluster sizes and depth

Per-cell QC metrics

⭑ canonical QC metric.

Notes

• no defects detected

• Cells: 6122
• Clusters: 14

The embedding shows all 6,122 cells, so the figure is not subsampled. The cells are split into 14 clusters of very uneven size. Cluster 0 is the largest, with 2,271 cells (about 37% of the total). Together with cluster 1 (968 cells), the two largest clusters hold just over half the dataset (3,239 cells). Cluster sizes then fall off steadily through clusters 2 to 10, from 601 cells down to 142. The three smallest clusters are cluster 11 with 69 cells and clusters 12 and 13 with 43 cells each (about 0.7% apiece). The large clusters therefore fill most of the plot, while the smallest clusters appear as a few sparse points.

• Clusters (seurat_clusters): 14 — 14 reviewed, 0 AI-draft, 0 suggestion-only · reference: Homo sapiens (23 cell types)

Per-cluster notes

Cluster 0 — B cell (reviewed)

The largest cluster carries the full B-cell receptor signature: MS4A1 (CD20), CD19 and the Igα/Igβ signalling chains CD79A/CD79B, together with BANK1 and FCRLA. The reference overlap matches four canonical genes and agrees with the curated label. This is a strong call.

Cluster 1 — T cell (reviewed)

The TCR/CD3 complex genes CD3E and CD3D, together with IL7R, IL32 and the GIMAP family, mark resting or memory αβ T cells. The reference overlap agrees on four genes. IL7R and KLRB1 are high and no CD8A appears among the top markers, which suggests a mostly CD4+ or memory-like population. A reviewer could refine the label with CD4 and CD8A expression.

Cluster 2 — Monocyte (reviewed)

Flag for reviewer: the marker evidence does not support this label. The top markers are epithelial and squamous or keratinocyte-type genes: the cytokeratins KRT17 and KRT7, S100A7, S100A2 and S100P, and the secretory or inflammatory epithelial genes SLPI, LCN2 and SAA1. In an NSCLC sample, this fits epithelial or tumour cells. The "Monocyte" call rests on a single gene, S100A9, with a reference-overlap score of 0.117. S100A9 is also induced in inflamed epithelium, and none of the monocyte genes that define cluster 4 (LYZ, FCN1, CD14) appear here. The curated reference has no epithelial type, so the reference could not offer one. The label should be checked against EPCAM, KRT8/KRT18 and CD14 before this cluster is carried into downstream analyses.

Cluster 3 — CD8+ T cell (reviewed)

The co-receptor CD8A and the TCR gene CD3E identify cytotoxic T cells. GZMK, GZMA, CST7, CTSW and NKG7 add a granzyme-K–high effector-memory profile, and CCL5 and CCL4 are consistent with it. The reference overlap agrees on three genes. NK-lineage genes (NKG7, KLRB1) give a minor second suggestion, which is expected for cytotoxic T cells.

Cluster 4 — Monocyte (reviewed)

LYZ, FCN1, S100A12, CD14 and VCAN are the canonical classical-monocyte genes (CD14+). TYROBP, AIF1 and CST3 support a myeloid identity. The reference overlap agrees on four genes, so this is a strong call.

Cluster 5 — Macrophages (reviewed)

No reference suggestion was made and no evidence genes are recorded, but the markers support a macrophage identity. APOE, APOC1, PLIN2 and CTSL are lipid-handling and lysosomal genes, SPP1 is typical of tumour-associated macrophages, and CCL2 and CXCL8 are monocyte-derived inflammatory chemokines. The reference set appears not to include a macrophage type, so the call stands on the markers alone. A reviewer should record evidence genes (for example APOE, APOC1, SPP1, together with CD68 or C1QA if expressed).

Cluster 6 — NK cell (reviewed)

GNLY, NKG7, GZMB and PRF1 form the NK cytotoxic-granule programme. KLRD1 (CD94) and XCL2 point to NK cells rather than cytotoxic T cells. The reference overlap agrees on four genes. Checking for absent CD3E would rule out an NKT or CD8 contribution.

Cluster 7 — Dendritic cell (reviewed)

High MHC class II expression (HLA-DQB2, HLA-DQA1, HLA-DPA1, HLA-DPB1) together with FCER1A and CST3 fits a conventional dendritic cell, most likely cDC2. The reference overlap supports the label on two genes. S100B and HLA-DQB2 can also mark Langerhans-like DCs, which a reviewer could check with CD1A and CD207.

Cluster 8 — Regulatory T (Treg) cell:Liver (reviewed)

FOXP3 is the Treg master transcription factor. CTLA4, TNFRSF4 (OX40), TNFRSF18 (GITR) and BATF complete an activated, tumour-infiltrating Treg profile, so the identity is well supported even though the curated reference made no suggestion and no evidence genes are recorded. Flag for reviewer: the ":Liver" suffix looks like a tissue tag inherited from a reference library. It describes anatomy, not identity, and does not match an NSCLC sample, so the label should probably read "Regulatory T cell". FOXP3, CTLA4 and TNFRSF18 should be recorded as evidence.

Cluster 9 — B Cells (reviewed)

Flag for reviewer: the markers point to plasma cells rather than B cells. The top markers are all immunoglobulin genes (IGHA1, IGKC, and IGKV, IGHV and IGLV variable segments) plus JCHAIN. This is the antibody-secreting programme of IgA plasma cells. Cluster 0's B-cell genes (MS4A1, CD19) are absent from the top markers. A more specific label, such as "Plasma cell", would also avoid a near-duplicate of cluster 0's "B cell" name. No evidence genes are recorded, and JCHAIN, IGHA1 and MZB1 or SDC1 (if expressed) would serve.

Cluster 10 — Mast Cells (reviewed)

The tryptases TPSAB1, TPSB2 and TPSD1, the carboxypeptidase CPA3, and MS4A2 (the FcεRI β chain) are canonical mast-cell genes, and HPGDS fits too. The identity is unambiguous even without a reference suggestion. A reviewer should record TPSAB1, CPA3 and MS4A2 as evidence.

Cluster 11 — CD8+ T cell (reviewed)

Flag for reviewer: this is a mixed signature. GZMK and KLRG1 fit effector-memory cytotoxic T cells. However, the B-cell genes MS4A1, CD79A and BANK1 are also among the top markers, and the recorded evidence is a single gene (GZMK). With only 69 cells, the cluster may hold T–B doublets or a small mixed population rather than a CD8+ subset. CD8A, CD3E and doublet scores should be checked before the label is relied on.

Cluster 12 — Plasmacytoid DC (reviewed)

LILRA4 and IRF7, recorded as evidence, are defining genes for plasmacytoid dendritic cells. PTGDS, TCL1A, SERPINF1, CLIC3 and GZMB complete the typical pDC profile. This is a strong call despite the small size of 43 cells.

Cluster 13 — Proliferating cell (reviewed)

STMN1, TUBA1B, TUBB, HMGB2, UBE2C, TYMS, HIST1H4C and H2AFZ are cell-cycle, DNA-replication and mitotic genes. MKI67 is recorded as evidence, although it is not among the top eight markers. The label describes a cell-cycle state, not a lineage. A reviewer could find the underlying type (for example cycling T cells or tumour cells) from lineage markers inside the cluster.

• Genes shown: 60 (top 5 per cluster)
• Clusters: 14
• Encoding: dot plot — colour = mean expression (normalized per gene), size = % of cells expressing

The figure shows the top-ranked marker genes for each of the 14 clusters. Each gene's mean expression is scaled to its own range, so colour shows which cluster expresses a gene most, not how highly that gene is expressed compared with other genes. Dot size shows how widely the gene is expressed within the cluster. Several genes in the panel have well-known cell-type associations. MS4A1 and CD79A are consistent with a B-cell identity. FOXP3, alongside TNFRSF4 and TNFRSF18, is consistent with regulatory T cells. The tryptase genes TPSAB1 and TPSB2, together with CPA3, are consistent with mast cells. These associations support the current cluster labels but do not confirm them on their own. A reviewer should read them together with the full marker tables.

Minimum spanning tree over 14 cluster centroids in PCA (15 dimensions), rooted at 13 · Proliferating cell. Pseudotime is distance from that root along the tree; each cell is placed between its cluster and the neighbour nearer the root.

A minimum spanning tree connects every cluster whether or not a lineage exists, so this figure shows which centroids are nearest each other in PCA space. It is not evidence that any population arises from another. The root is not a finding either. The module picked cluster 13 by its default rule, "the most peripheral leaf of the tree", and nobody supplied a known origin. The direction of pseudotime is therefore an assumption: rooting the same tree at cluster 5 would reverse the axis. Treat the cell-type names as provisional. They come from the identities layer, and several look like reference-library suggestions rather than curated labels: "Regulatory T (Treg) cell:Liver" carries a tissue suffix from a reference, and "B cell" (0) and "B Cells" (9) are spelled differently. Cluster 0 is called "B cell" yet sits 0.65x the median spread from cluster 11, "CD8+ T cell", which is closer than any T-cell pair in the table. Either those labels or that proximity needs checking before the tree is read biologically.

Backbone

Only one part of the tree is tightly connected: a lymphoid core in which clusters 0, 11, 1, 3, 8 and 6 are joined by edges of 0.65x–1.27x the median cluster spread (6.94). Every other cluster hangs off that core by a long edge. Five edges exceed 3x the spread and are weak links, not neighbourhoods: 11–7 (3.89x), 11–12 (4.85x), 1–2 (4.92x), 11–10 (5.01x) and 8–13 (6.27x). The defects note lists only four of them and leaves out 8–13, which is the longest edge in the tree and the one that attaches the root. Edge 7–4 sits exactly at 3.0x. Because of the 8–13 edge, every non-root cluster starts at pseudotime ≥ 43.489 (cluster 8), so nearly half of the axis before the rest of the tree begins is a single gap in PCA space. Past that gap, the pseudotime ordering mostly records distance from the lymphoid core. The T and NK clusters come earliest (1 at 51.346, 11 at 56.351, 3 at 56.888, 6 at 65.72). The myeloid clusters come latest (2 at 85.442, 4 at 104.117, 5 at 122.976), with 12 (89.982) and 10 (91.112) branching off separately.

Pseudotime by cluster

Genes changing along pseudotime

All six top genes rise with pseudotime, from SOD2 (r = +0.86) to LILRB4 (r = +0.82). All six are myeloid genes: FCER1G and TYROBP rise from 0.666 and 0.824 to about 3.4, S100A9 from 0.243 to 1.993, and FTL from 2.561 to 5.628. Their profiles read zero in bins 2–3, the stretch covered only by the long 13–8 edge. They increase most steeply in the last three bins, where monocyte clusters 4 and 2 and macrophage cluster 5 sit. The simplest reading is that these correlations record which cell types occupy the late end of the axis. They are not a maturation programme unfolding within one population. On known biology, the evidence is mixed:

• Consistent: macrophage cluster 5 is placed next to and after monocyte cluster 4 (edge at 2.72x), which matches the known monocyte-to-macrophage relationship. The tree ordering is consistent with that relationship but does not demonstrate it.
• Contrary: the axis runs from a proliferating cluster through T cells to myeloid cells. That does not match any known relationship between lymphoid and myeloid cells in tissue.
• Contrary: dendritic cells are placed before monocytes (7 → 4), which reverses the usual monocyte-to-DC direction.
• Unresolved: a proliferating root upstream of everything would suit a progenitor, but in tumour-infiltrating immune cells a cycling cluster is more often proliferating T or NK cells. Its 6.27x edge says it is not close to anything.

Nothing here observed time or tracked a cell from one state to another. This section supports only statements about proximity in PCA space. A lineage claim would need a biologically chosen root (the root parameter) and an independent method such as RNA velocity or lineage tracing.

Notes

• ⚠ a minimum spanning tree always returns a tree — these clusters are connected because the algorithm connects everything, not because a lineage was detected. Read the edge lengths against the median cluster spread (6.94) before believing any branch
• ⚠ root: most peripheral leaf of the tree — ARBITRARY, set the root param when the origin is known. Pseudotime is distance from that root, so the direction of the axis is a choice, not a finding
• ⚠ 5 edge(s) are more than 3x the median cluster spread (11–7, 11–12, 1–2, 11–10) — those clusters are probably not neighbours in any developmental sense

Top 25 markers per cluster tested against 920 WikiPathways gene sets for Homo sapiens (968 in the index; sets with <3 measured genes dropped). Background = the 7304 pathway genes measured in this dataset; hypergeometric over-representation, Benjamini-Hochberg within each cluster; fold = (k/n)/(K/N).

12 of 14 clusters have at least one hit; 14 hit(s) rest on the 3-gene overlap floor and are marked †; 40 pathway(s) matching the same genes as a listed hit were folded into it. Markers are UP-regulated genes, so a pathway coherently down in a cluster cannot appear.

Several lymphoid clusters show a pathway signal that fits their labels. B cell cluster 0 hits memory B cell differentiation (4/46, 52.93x, q=0.00042; CD79A, CD79B, IRF8, POU2AF1), which the index also lists as B cell receptor signaling. T cell cluster 1 hits TCR signalling modulators (4/61, 39.91x, q=0.0019). In CD8+ T cell cluster 3 the top hit carries a SARS-CoV-2 title, but its genes (CD2, CD3E, CD8A, CD8B, FYN; 5/23, 158.78x, q=4.5e-08) make it a T-cell receptor programme, and it is also listed as T-cell receptor signaling and as PD-1 blockade. NK cluster 6 has a cytotoxic set (CCL4, CCL5, GZMB, IFNG, PRF1; 5/27, 122.96x, q=2e-07). Treg cluster 8 has a checkpoint and co-stimulation set (CTLA4, ICOS, TIGIT, TNFRSF4; 4/54, 49.19x, q=0.00078). Among the myeloid and cycling clusters, dendritic cell cluster 7 gives the strongest result in the dataset: Allograft rejection, 10/79, 54.39x, q=2.1e-13. That hit comes from MHC class II genes plus C1QA/B/C. Class II fits antigen-presenting cells, but C1Q is more typical of macrophages than dendritic cells. Monocyte cluster 4 hits complement and phagocytosis (C1QA, C1QB, C1QC, FCER1G, TYROBP; 5/40, 57.06x, q=1.5e-05). Macrophage cluster 5 hits lung fibrosis (CCL2, CXCL8, SPP1, TIMP1; 4/61, 29.93x, q=0.0068). Proliferating cluster 13 hits Retinoblastoma gene in cancer, which here reads as an E2F/S-phase programme (TOP2A, RRM2, TYMS, CCNB2; 8/87, 39.51x, q=6e-09).

Some rows do not support a finding, and some raise questions about the label. The following hits have q ≥ 0.05 and are not supported after multiple-testing correction, even where the name sounds apt:

• in cluster 1: cytokine-cytokine receptor interaction and the SARS-CoV-2 network (q=0.35, 0.57)
• in cluster 2: vitamin D receptor (q=1)
• in cluster 3: cytokine-cytokine receptor interaction (q=0.11)
• in cluster 4: spinal cord injury and vitamin D receptor (q=0.13, 0.28)
• in cluster 5: nuclear receptors (q=0.57)
• in cluster 13: SASP and Cell cycle (q=0.1, 0.26)

Every hit in cluster 9 and the top hit in cluster 11 rest on the 3-gene floor, so the large folds there are weak evidence. In cluster 9 both † hits are made of immunoglobulin genes (IGHG1, IGHG2, IGLC2; IGHA1, IGHA2, IGHM). That points to antibody-secreting cells more than to the "B Cells" label or to the pathway titles. Cluster 11 is labelled CD8+ T cell, yet its only hit is a B-lineage set (CD79A, CD79B, TCF4; 3/46, q=0.025), and MS4A1 is among its markers. The hit is weak, but the identity is worth checking. Cluster 2 is labelled Monocyte, but its highly significant hit is a keratinocyte/squamous set (KRT17, KRT6A, KRT7, SERPINB3, SERPINB4; 6/45, 54.1x, q=6.3e-07). Its uncovered markers (ELF3, PERP, AQP3, LCN2) are epithelial as well, which points to epithelial or tumour cells rather than monocytes. The ":Liver" suffix on cluster 8 looks like a reference-transfer artefact. None of these labels carries a trailing "?", but they may still be suggestions rather than reviewed identities, and they should be reviewed before any row is read against them. Clusters 10 (Mast Cells) and 12 (Plasmacytoid DC) have no reported hit.

The caveats below apply to every row:

• Background: the test's universe is only the 7304 pathway-annotated genes measured in this dataset, so fold and p describe enrichment relative to curated genes, not the whole transcriptome.
• Up-regulated genes only: the markers include only up-regulated genes, so nothing here says a programme is reduced anywhere.
• Per-cluster correction: BH was applied separately in each of 14 families of 920 tests, so no single q is corrected across the dataset.
• Cells, not replicates: the markers count cells, not replicates, so these are candidate programmes rather than measured effects. Overlap with a curated set is suggestive, not causal.
• Pathway titles: WikiPathways titles often name the disease in which a gene set was curated (SARS-CoV-2, pancreatic cancer, allograft rejection, retinoblastoma) rather than anything happening in these lung-tumour cells. Read the overlapping genes, not the title.

Top hits per cluster

Cluster 0 — B cell

Markers in no reported pathway: AIM2, BLK, CD19, CD37, CD74, HLA-DQA1, MS4A1, TNFRSF13B.

Cluster 1 — T cell

Markers in no reported pathway: FYB, LDHB, LEF1.

Cluster 2 — Monocyte

Markers in no reported pathway: AQP3, CD55, ELF3, FABP5, GSTP1, LCN2, PERP, S100P, SAA1, SLPI.

Cluster 3 — CD8+ T cell

Markers in no reported pathway: CD3D, KLRG1.

Cluster 4 — Monocyte

Markers in no reported pathway: CST3, FCN1, FPR1, LYZ, MAFB, SERPINA1.

Cluster 5 — Macrophages

Markers in no reported pathway: APOC1, APOE, CD68, CTSB, CTSL, FCER1G, G0S2, IL1RN, PLAUR, SOD2.

Cluster 6 — NK cell

Markers in no reported pathway: CD247, CD8A, GNLY, KIR2DL4, XCL1, XCL2.

Cluster 7 — Dendritic cell

Markers in no reported pathway: ALDH2, CLEC10A, CST3, FCER1A, GSN, HLA-DQB2, S100B.

Cluster 8 — Regulatory T (Treg) cell:Liver

Markers in no reported pathway: BATF, CD3D, PIM2, SIRPG, TNFRSF18.

Cluster 9 — B Cells

Cluster 11 — CD8+ T cell

Markers in no reported pathway: CD40LG, FYB, KLRG1, LTB, MAP4K1, MS4A1, TUBA4A.

Cluster 13 — Proliferating cell

Markers in no reported pathway: HIST1H4C, MKI67, TUBA1B, TUBB.

Notes

• ⚠ this tests UP-regulated markers only — cluster_markers.json holds positively-scored genes — so a pathway coherently DOWN in a cluster cannot appear here at all
• ⚠ Benjamini-Hochberg is applied WITHIN each cluster, so these are 14 separate families of 920 tests; a q<0.05 in one of them has had 14 families' worth of opportunity
• ⚠ the marker lists this rests on come from a test that counts CELLS, not replicates, so the enrichment inherits that: these are candidate programmes to follow up, not an effect measured across animals
• ⚠ 14 hit(s) rest on exactly 3 shared gene(s), the overlap floor — the weakest claim this module makes, marked † in the table
• ⚠ 40 further pathway(s) matched the same genes as a hit already listed and were folded into it as aliases; collapse_overlap=0 reports them separately

Top 25 markers per cluster tested against 4559 GO terms for Homo sapiens (4618 in the index; terms with <3 measured genes dropped). Background = the 12669 GO-annotated genes measured in this dataset; hypergeometric over-representation, Benjamini-Hochberg within each cluster; fold = (k/n)/(K/N). GO is a DAG, so terms whose gene sets largely repeat a stronger hit are folded into it and listed as also.

The clearest programmes are antigen presentation, cytotoxicity and antibody production. In cluster 7, 8 of the 14 genes in peptide antigen assembly with MHC class II protein complex are markers (381.02x, q 6.3e-17), all of them HLA class II chains. That is the best-supported hit in the dataset because both the fold and the coverage of a small term are high. Cluster 6 carries the cytotoxic machinery (GZMA, GZMB, NKG7, PRF1) under granzyme-mediated programmed cell death signaling pathway (4/8, 316.73x, q 1.4e-06). It also carries inhibitory NK receptors (KIR2DL4, KLRC1, KLRD1) under negative regulation of natural killer cell mediated cytotoxicity (118.77x, q 0.0017), which is consistent with its NK label. Cluster 3 matches the same granzyme term with three genes (3/8, 237.54x, q 0.00039) and T cell receptor signaling pathway (34.43x, q 0.00039) with CD3D/CD3E/CD8A/CD8B, which reads as cytotoxic CD8+ T cells, as labelled. Cluster 0's B cell receptor signaling pathway (6/50, 95.02x, q 9.8e-08) and cluster 13's DNA topological change (HMGB1, HMGB2, TOP2A; 248.41x, q 0.00076) also agree with their labels. Three caveats apply to these readings:

• Small terms and few genes. A fold above 200x on an 8- or 9-gene term rests on three or four genes, so it should be read as "these specific genes are present", not as a pathway-wide signal.
• Redundant ancestors. The "+N redundant" notes (+5 on clusters 7 and 9) are the same genes repeated under ancestor terms in the GO DAG. They are annotation structure, not independent confirmation.
• Broad terms. Cluster 1's top term, cell surface receptor signaling pathway (7/301, 16.37x), is a large umbrella term, and the same term appears in clusters 0 and 3. The more specific positive regulation of T cell proliferation (4/48, 58.65x, q 0.0012) says more about cluster 1 despite its larger q.

Several labels are not supported by their own terms, and all of these labels should be treated as provisional until reviewed.

• Cluster 11 (labelled CD8+ T cell) enriches only for B cell receptor signaling pathway (4/50, 72.39x, q 0.00095) through BANK1, CD79A, CD79B and MS4A1. That is a B-lineage signature. The label needs checking against the marker table, including the possibility of B/T doublets.
• Cluster 2 (labelled Monocyte) hits endothelial cell migration (5/33, 95.98x, q 6e-06). The term name comes from how S100A2/A7/A8/A9/P are annotated in GO, and it says nothing about endothelium. Together with KRT17/KRT6A/KRT7 under intermediate filament organization (41.31x, q 0.048, only just under the 0.05 cutoff), the markers look more epithelial (squamous-like) than monocytic.
• Cluster 4 (labelled Monocyte) combines the complement C1q genes (C1QA/B/C; synapse pruning, 222.26x, q 0.0011) with S100A8/A9/A12 (defense response to fungus, 142.88x, q 0.0016). Again the term names describe annotation, not tissue. C1q is more typical of macrophages, so this may be a mixed myeloid cluster.
• Cluster 7 (Dendritic cell) also carries the C1q triplet (222.26x, q 0.00018), which is worth checking before the DC label is signed.
• Cluster 9 (B Cells) tops out at antibacterial humoral response (7/37, 199.74x, q 3.6e-12). Its genes are immunoglobulin constant chains and JCHAIN, which points to antibody-secreting plasma cells rather than B cells in general.

Clusters 5, 8, 10 and 12 have no term at q < 0.05. That means only that their top markers do not concentrate in any GO process; it is neither evidence for nor against their labels. This section tests marker lists only and makes no claim about differences between conditions.

Terms per cluster

Cluster 0 — B cell

Cluster 1 — T cell

Cluster 2 — Monocyte

Cluster 3 — CD8+ T cell

Cluster 4 — Monocyte

Cluster 6 — NK cell

Cluster 7 — Dendritic cell

Cluster 9 — B Cells

Cluster 11 — CD8+ T cell

Cluster 13 — Proliferating cell