GSE274058 Reference Results

这页发布的是 shen_2026_scrnaseq (GSE274058) 的 dissociated reference-side 结果包。它服务于 reference signature、program reconstruction 和后续 cross-platform concordance；它不是 spatial benchmark 的主图页面。

Provenance

  • Dataset card: shen_2026_scrnaseq (GSE274058).

  • Result type: dissociated reference-side pseudobulk intrinsic DE package.

  • Generated at (UTC): 2026-04-28T18:47:25.610270+00:00.

  • Duration: 365.981 seconds.

  • Command: scripts/run_gse274058_reference.py.

  • Python: 3.10.12 on Linux-5.15.0-1072-nvidia-x86_64-with-glibc2.35.

  • SpatialPerturb version: 0.3.0.

  • Source report directory: /data/taobo.hu/SpatialPerturb_a100_release_rerun/reports/gse274058_reference_run.

GitHub Release Assets

A100 Confirmation

  • Status: replaced.

  • Baseline report: D:\GitHub\SpatialPerturb\reports\gse274058_reference_run.

  • Candidate report: D:\GitHub\SpatialPerturb\reports\a100_gse274058_reference_run.

  • Compared at (UTC): 2026-04-28T18:48:11.160414+00:00.

  • Authoritative source: /data/taobo.hu/SpatialPerturb_a100_release_rerun/reports/gse274058_reference_run.

  • Outcome: A100 rerun replaced the local draft package for release.

当前 public summary 已按 A100 重跑结果更新；若后续 A100 rerun 与本页不同，release assets 会继续以最新权威 rerun 为准重新生成。

Dataset Scale and Barcode QC

  • Cells: 41749

  • Genes: 32317

  • Samples: 5

  • Barcode status counts: {'multiple': 285, 'single': 707, 'unassigned': 40757}

  • Single-cell control count: 24

  • Valid perturbations: 17

  • Successful perturbations: 17

Barcode spread

Valid Perturbations

下表把 inference QC 通过的 perturbations 与一个简化版 knockdown_adequate 标记放在一起，便于在论文叙事里区分“能算”与“值得主张”的层级。

perturbation

n_cells

fraction_cells

target_gene

target_log2fc

knockdown_adequate

Clu

80

0.113

Clu

-0.880

yes

Stk39

65

0.092

Stk39

-0.612

yes

Rraga

60

0.085

Rraga

-0.136

yes

Gfap

55

0.078

Gfap

0.134

no

Fasn

49

0.069

Fasn

-0.079

yes

Flcn

46

0.065

Flcn

0.061

no

Rbfox3

46

0.065

Rbfox3

0.000

no

Dpp6

40

0.057

Dpp6

0.104

no

Trem2

40

0.057

Trem2

0.692

no

C9orf72

37

0.052

C9orf72

0.060

no

Cfap410

30

0.042

Cfap410

0.047

no

Tbk1

28

0.040

Tbk1

-0.030

yes

Olig2

27

0.038

Olig2

0.552

no

Sh3gl2

24

0.034

Sh3gl2

0.047

no

Lrrk2

23

0.033

Lrrk2

0.177

no

Srf

19

0.027

Srf

0.085

no

Ndufaf2

14

0.020

Ndufaf2

-0.155

yes

Focus Perturbations

当前结果优先展示 Lrrk2 和 Srf，因为它们都是真实跑出来的 valid perturbations，同时也最能说明 reference-side signatures 与 target knockdown 质量之间并不总是等价。

perturbation

gene

log2fc

fdr

case_n

control_n

Lrrk2

Ttc39c

-2.000

3.705e-04

23

24

Lrrk2

AI597479

-1.737

3.705e-04

23

24

Lrrk2

Zfhx2os

-2.322

0.0063

23

24

Lrrk2

Tex52

-2.000

0.0063

23

24

Lrrk2

Usp13

-1.585

0.00783

23

24

Lrrk2

Sh3gl2

-1.585

0.00783

23

24

Lrrk2

H2afx

-1.585

0.00783

23

24

Lrrk2

Plk3

-2.000

0.00879

23

24

Lrrk2

Hist1h4h

-1.874

0.00879

23

24

Lrrk2

Gm28322

-1.585

0.00879

23

24

Srf

Bcdin3d

-1.737

1.919e-05

19

24

Srf

Mrpl10

-1.585

0.00107

19

24

Srf

Gm28322

-1.585

0.00107

19

24

Srf

Ldlr

-1.585

0.00107

19

24

Srf

Atxn3

-1.585

0.00107

19

24

Srf

Sephs1

-1.415

0.00107

19

24

Srf

Vwa1

-1.415

0.00107

19

24

Srf

Vps25

-1.415

0.00107

19

24

Srf

Mettl3

-1.415

0.00107

19

24

Srf

N4bp2

-1.737

0.00165

19

24

Target-Gene Sanity Check

perturbation

gene

log2fc

fdr

mean_case

mean_control

Lrrk2

Lrrk2

0.807

1

0.750

0.000

Srf

Srf

0.070

1

0.400

0.333

Program Summary

program_matrix.tsv 已覆盖全部成功跑完 pseudobulk intrinsic DE 的 perturbations。这里展示每个 perturbation 对应的 program gene count，方便检查后续 cross-platform 对齐时的覆盖范围。

perturbation

program_gene_count

C9orf72

50

Cfap410

50

Clu

50

Dpp6

50

Fasn

50

Flcn

50

Gfap

50

Lrrk2

50

Ndufaf2

50

Olig2

50

Rbfox3

50

Rraga

50

Sh3gl2

50

Srf

50

Stk39

50

Tbk1

50

Trem2

50

How To Improve

  • Target knockdown quality is unstable: only 6 of 17 valid perturbations show target_log2fc < 0, and neither Lrrk2 nor Srf shows a convincing target-gene decrease in this run. Add a knockdown_adequate QC flag and separate weak perturbations from main claims.

  • Sample and cell balance is tight: control has only 24 single cells across 3 samples, and several perturbations sit between 14 and 23 cells. Add per-sample minimum cell thresholds and stricter sample-level exclusion rules before treating reference signatures as stable.

  • The raw DE table is large for routine browsing (intrinsic_de.tsv is about 52.7 MB). Keep publishing compressed .tsv.gz assets by default, and consider adding .parquet export for downstream reuse.

  • Documentation is still dual-stack today: RTD uses Sphinx while local site configuration still carries MkDocs nav. This publish keeps Sphinx for stability, but the long-term maintenance path should converge to one docs stack.