Upload outputs/runs/single_085_090_human_only_20260306_032339/subset090/reports/confound/gnn_confound_subset090.md with huggingface_hub
Browse files
outputs/runs/single_085_090_human_only_20260306_032339/subset090/reports/confound/gnn_confound_subset090.md
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Mesh vs GSE Confound Check
|
| 2 |
+
|
| 3 |
+
- embedding: `/home/zzz0054/deliverables/data/human_only_085_090/subset090/gnn_old_holdout_subset_090_human.npz`
|
| 4 |
+
- enriched metadata: `/home/zzz0054/deliverables/data/human_only_085_090/subset090/enriched_gs_metadata_subset090_human.csv`
|
| 5 |
+
- seed: 11 | k: 10 | N: 1626
|
| 6 |
+
- min_support: 5 | fdr_q: 0.05
|
| 7 |
+
|
| 8 |
+
## Confound Association (Primary Labels)
|
| 9 |
+
|
| 10 |
+
- primary GSE: items=1626 | unique=406 | NMI=0.351 | ARI=0.018
|
| 11 |
+
- primary platform: items=1626 | unique=23 | NMI=0.054 | ARI=0.007
|
| 12 |
+
|
| 13 |
+
## Cluster Summary
|
| 14 |
+
|
| 15 |
+
| cluster | size | top_gse | gse_purity | mesh_sig | best_mesh (q,ER) | gse_sig | best_gse (q,ER) | plat_sig | best_plat (q,ER) |
|
| 16 |
+
|---:|---:|---|---:|---:|---|---:|---|---:|---|
|
| 17 |
+
| 0 | 160 | GSE213142 | 0.087 | 3 | Transposable (1.4e-03,7.11) | 2 | GSE242259 (1.1e-04,10.16) | 1 | "GPL24676" (3.4e-03,1.56) |
|
| 18 |
+
| 1 | 258 | GSE213142 | 0.043 | 22 | CD8-Positive (3.1e-07,3.69) | 4 | GSE159759 (1.4e-03,6.30) | 1 | "GPL15520" (4.6e-03,5.40) |
|
| 19 |
+
| 2 | 138 | GSE218585 | 0.080 | 20 | STAT1 (7.8e-07,8.64) | 2 | GSE218585 (2.7e-07,8.64) | 0 | |
|
| 20 |
+
| 3 | 130 | GSE214921 | 0.046 | 22 | Male (1.2e-10,2.90) | 2 | GSE214921 (3.7e-05,12.51) | 2 | "GPL20301" (5.4e-03,2.38) |
|
| 21 |
+
| 4 | 148 | GSE213142 | 0.108 | 7 | Proteins (3.0e-04,1.59) | 1 | GSE80963 (4.2e-02,7.32) | 0 | |
|
| 22 |
+
| 5 | 171 | GSE219208 | 0.111 | 9 | Host-Pathogen (1.1e-03,6.34) | 2 | GSE219208 (9.6e-17,9.51) | 2 | "GPL21697" (6.8e-06,3.57) |
|
| 23 |
+
| 6 | 219 | GSE216249 | 0.123 | 30 | Liposomes (5.5e-13,5.42) | 6 | GSE216249 (2.3e-14,5.42) | 4 | GPL15520 (7.7e-11,4.24) |
|
| 24 |
+
| 7 | 131 | GSE213142 | 0.092 | 5 | RNA-Seq (1.2e-02,3.16) | 0 | | 0 | |
|
| 25 |
+
| 8 | 127 | GSE220620 | 0.063 | 1 | Tumor (4.7e-02,1.59) | 1 | GSE220620 (4.3e-02,3.66) | 0 | |
|
| 26 |
+
| 9 | 144 | GSE213142 | 0.139 | 12 | Dose-Response (8.3e-03,6.77) | 1 | GSE213142 (1.5e-02,2.30) | 0 | |
|
| 27 |
+
|
| 28 |
+
## Interpretation Guide
|
| 29 |
+
|
| 30 |
+
- High `mesh_sig` with moderate/low `gse_purity` suggests biology-like themes not tied to one study.
|
| 31 |
+
- Very high `gse_purity` and strong `best_gse` indicates a study/batch-driven cluster (still real structure, but a confound for cross-study biology).
|
| 32 |
+
- Platform enrichment is a common technical confound; treat strong `best_plat` similarly.
|