Dataset Viewer
Auto-converted to Parquet Duplicate
cell
string
site
string
modality
string
is_primary
bool
verdict
string
carrying_axis
int64
carrying_axis_var_frac
float64
neg_eig_mass
float64
rho_comp
float64
p_comp
float64
p_comp_bonferroni9
float64
partial_rho
float64
partial_p
float64
rho_coverage
float64
rho_star_uniq
float64
est_coverage_cor
float64
kw_chem2v4_p
float64
adonis_r2_estimatescore
float64
adonis_p_estimatescore
float64
adonis_r2_coverage
float64
betadisper_coverage_p
float64
betadisper_estimatescore_p
float64
tpm_top2000_sig_overlap
int64
psi_host_overlap_genes
int64
psi_host_overlap_events
int64
excl_rho_comp
float64
excl_partial_rho
float64
excl_n_events_dropped
int64
Ome_PSI
Ome
PSI
true
composition_axis
1
0.239255
0
-0.697442
0.000018
0.000166
-0.632966
0.000395
-0.226366
0.174638
0.135507
0.00051
0.051061
0.0401
0.050316
0.0776
0.5118
null
4
4
-0.697442
-0.632966
4
Ome_TPM
Ome
TPM
false
composition_axis
2
0.14425
0
-0.759288
0.000001
0.00001
-0.758129
0.000005
-0.424157
-0.114127
0.135507
0.000043
0.070833
0.0152
0.030537
0.2469
0.6282
43
null
null
null
null
null
Asc_PSI
Asc
PSI
true
composition_axis
1
0.172002
0
-0.841157
0
0
-0.899036
0
-0.07851
0.02158
-0.218803
0.000968
0.112837
0.0001
0.040225
0.8712
0.6438
null
5
7
-0.841157
-0.899036
7
Asc_TPM
Asc
TPM
false
composition_axis
1
0.287134
0
-0.958621
0
0
-0.948615
0
0.253151
0.135929
-0.218803
0.172625
0.219686
0.0001
0.041465
0.6807
0.0017
169
null
null
null
null
null
Per_PSI
Per
PSI
true
composition_axis
1
0.15013
0
-0.734372
0.000004
0.000035
-0.699559
0.000049
-0.224254
0.182202
-0.041908
0.000042
0.073076
0.0002
0.036497
0.7334
0.3506
null
3
3
-0.734372
-0.699559
3
Per_TPM
Per
TPM
false
composition_axis
2
0.136819
0
0.78198
0
0.000003
0.707298
0.000037
-0.028533
-0.096774
-0.041908
0.012014
0.118785
0.0003
0.037296
0.2367
0.9684
61
null
null
null
null
null

within-site-subtypes-composition-v1

Within-site (Ome/Asc/Per, PDS, n=30/site) continuous-axis-vs-composition test. RESULT (confirmatory, prereg ec64515 + Amendment 1 72da6ff): all 3 PRIMARY PSI cells + all 3 TPM cells = composition_axis. The dominant continuous axis of within-site splicing (PSI) and expression (TPM) variation is a tumor-composition (ESTIMATEScore) gradient: |rho| 0.70-0.84 (PSI), survives partial control for coverage+depth+chemistry, beats coverage (est_cov_cor 0.04-0.22, not collinear), adonis R2_comp>=R2_cov, betadisper clean, and SURVIVES exclusion of the 3-5 ESTIMATE-282-hosted events (non-circular). Explains the parent's exploratory Per/Asc cross-modal concordance.

Dataset Info

  • Rows: 6
  • Columns: 28

Columns

Column Type Description
cell Value('string') site_modality identifier
site Value('string') No description provided
modality Value('string') No description provided
is_primary Value('bool') True for the 3 PSI cells (least circular; PSI never contains ESTIMATE gene expression)
verdict Value('string') composition_axis / technical_depth_axis / no_dominant_axis / ambiguous_collinear (locked rule)
carrying_axis Value('int64') which of the top-3 PCoA axes carries composition (argmax
carrying_axis_var_frac Value('float64') that axis's eigenvalue fraction (sum
neg_eig_mass Value('float64') No description provided
rho_comp Value('float64') Spearman rho(composition axis, ESTIMATEScore) — the primary effect
p_comp Value('float64') No description provided
p_comp_bonferroni9 Value('float64') p_comp x 9 (3 axes x 3 scores Bonferroni; sig < 0.05)
partial_rho Value('float64') partial Spearman rho(axis, ESTIMATEScore
partial_p Value('float64') No description provided
rho_coverage Value('float64') rho(axis, psi_median_coverage) — the technical rival
rho_star_uniq Value('float64') No description provided
est_coverage_cor Value('float64') Spearman cor(ESTIMATEScore, coverage);
kw_chem2v4_p Value('float64') No description provided
adonis_r2_estimatescore Value('float64') vegan::adonis2 marginal R2 for ESTIMATEScore (by=margin)
adonis_p_estimatescore Value('float64') No description provided
adonis_r2_coverage Value('float64') vegan::adonis2 marginal R2 for coverage (rival)
betadisper_coverage_p Value('float64') betadisper permutest p over coverage tertiles; <0.05 caveats the adonis location term
betadisper_estimatescore_p Value('float64') No description provided
tpm_top2000_sig_overlap Value('int64') TPM cells:
psi_host_overlap_genes Value('int64') PSI cells:
psi_host_overlap_events Value('int64') No description provided
excl_rho_comp Value('float64') PSI cells: rho_comp after excluding ESTIMATE-282-hosted events (Amendment 1.2 sensitivity)
excl_partial_rho Value('float64') No description provided
excl_n_events_dropped Value('int64') No description provided

Generation Parameters

{
  "script_name": "23_composition_association.py (+ 20-22,24,25)",
  "model": "N/A \u2014 ESTIMATE (ssGSEA) + PCoA + partial Spearman + vegan::adonis2",
  "description": "Within-site (Ome/Asc/Per, PDS, n=30/site) continuous-axis-vs-composition test. RESULT (confirmatory, prereg ec64515 + Amendment 1 72da6ff): all 3 PRIMARY PSI cells + all 3 TPM cells = composition_axis. The dominant continuous axis of within-site splicing (PSI) and expression (TPM) variation is a tumor-composition (ESTIMATEScore) gradient: |rho| 0.70-0.84 (PSI), survives partial control for coverage+depth+chemistry, beats coverage (est_cov_cor 0.04-0.22, not collinear), adonis R2_comp>=R2_cov, betadisper clean, and SURVIVES exclusion of the 3-5 ESTIMATE-282-hosted events (non-circular). Explains the parent's exploratory Per/Asc cross-modal concordance.",
  "experiment_name": "within-site-subtypes",
  "experiment_id": "within-site-subtypes",
  "artifact_type": "analysis_result",
  "visualizer_type": "table",
  "cluster": "local",
  "artifact_status": "final",
  "canary": false,
  "hyperparameters": {
    "estimate_platform": "illumina (relative ESTIMATEScore, no absolute purity)",
    "ensg_to_symbol": "org.Hs.eg.db",
    "pcoa": "classical MDS top-3 (sum|lambda| denom)",
    "association": "Spearman + partial (| coverage,star_uniq,chem2v4)",
    "adonis": "vegan::adonis2 by=margin 9999 perm + betadisper",
    "seed": 20260713,
    "bonferroni": "3 axes x 3 scores = 9",
    "rho_floor": 0.5,
    "collinearity_downgrade": "|cor(EST,cov)|>=0.6"
  },
  "input_datasets": []
}

Usage

from datasets import load_dataset

dataset = load_dataset("depinwang/within-site-subtypes-composition-v1", split="train")
print(f"Loaded {len(dataset)} rows")

Downloads last month
17