# GSE293514 external human-myoblast CRISPR phenotype stress test

## Scope

This analysis intersects observed HepG2 CRISPRi DMD-rescue-like scores with the independent GSE293514 human-myoblast fusion CRISPR screen. The external endpoint is loss-of-function impairment of myoblast fusion, not a DMD-rescue transcriptome, so the analysis is a muscle-context phenotype and candidate-safety stress test rather than direct validation of therapeutic rescue.

## Coverage

| metric | value |
| --- | --- |
| published_screen_genes | 7197 |
| published_fusion_hits_fdr_lt_0_1 | 250 |
| eligible_hepg2_perturbations | 2160 |
| gene_overlap | 1575 |
| fusion_hits_in_overlap | 182 |
| top_rescue_decile_size | 158 |
| fusion_hits_in_top_rescue_decile | 13 |
| feasible_dmd_kd_queue_overlap | 273 |
| feasible_dmd_kd_queue_fusion_risk_flags | 9 |

## Primary statistical tests

| test | effect | ci95_low | ci95_high | p_value | interpretation | holm_p |
| --- | --- | --- | --- | --- | --- | --- |
| spearman_rescue_vs_fusion_lfc | -0.00775733 | -0.0557985 | 0.0419839 | 0.758372 | association only; neither variable is a causal endpoint | 0.758372 |
| mann_whitney_fusion_hit_vs_nonhit_rescue | -0.0532174 | -0.00968007 | 0.00231373 | 0.242403 | rank-biserial effect; CI columns are mean-difference CI (point -0.00382905) | 0.570819 |
| fisher_top_rescue_decile_vs_fusion_hit | 0.662069 |  |  | 0.190273 | odds ratio; external muscle-fusion safety stress test | 0.570819 |
| fisher_feasible_queue_vs_fusion_hit | 0.222478 |  |  | 2.14102e-07 | odds ratio; queue enrichment is not therapeutic validation | 8.56409e-07 |

Multiplicity-supported tests: fisher_feasible_queue_vs_fusion_hit.

## Candidate-level safety flags

| gene | rescue_score_minus_cosine | candidate_class | feasible_queue_rank | pos|fdr | pos|lfc | fusion_hit_fdr_lt_0_1 | external_muscle_fusion_flag |
| --- | --- | --- | --- | --- | --- | --- | --- |
| SRA1 | -0.00383517 | C_exploratory_kd_candidate | 4 | 0.999999 | -0.23874 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| CAST | 0.0325414 | C_exploratory_kd_candidate | 6 | 0.999999 | -0.12399 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| ADAM10 | 0.0165779 | C_exploratory_kd_candidate | 7 | 0.999999 | -0.075195 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| MEF2A | -0.0768127 | C_exploratory_kd_candidate | 8 | 0.999999 | -0.24002 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| PSME1 | 0.0358873 | C_exploratory_kd_candidate | 9 | 0.999999 | -0.11854 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| GIT2 | -0.0104928 | C_exploratory_kd_candidate | 10 | 0.319394 | 0.44243 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| GSK3B | -0.100602 | C_exploratory_kd_candidate | 11 | 0.999999 | -0.01275 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| THRAP3 | 0.012686 | C_exploratory_kd_candidate | 12 | 0.999999 | 0.044113 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| IST1 | -0.0109855 | C_exploratory_kd_candidate | 13 | 0.999999 | -0.075942 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| FBXW7 | -0.0524582 | C_exploratory_kd_candidate | 14 | 0.144687 | 0.78628 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| SMARCC1 | 0.0588742 | C_exploratory_kd_candidate | 15 | 0.999999 | 0.038593 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| ANKRD17 | 0.0101111 | C_exploratory_kd_candidate | 16 | 0.999999 | 0.061971 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| CALR | 0.034272 | C_exploratory_kd_candidate | 17 | 0.999999 | -0.065853 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| PSME2 | -0.0352475 | C_exploratory_kd_candidate | 18 | 0.999999 | -0.015033 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| NAA38 | -0.0347857 | C_exploratory_kd_candidate | 19 | 0.010091 | 1.6593 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |
| SPAG7 | -0.00480276 | C_exploratory_kd_candidate | 20 | 0.999999 | -0.04148 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| ITGB1BP1 | 0.11773 | C_exploratory_kd_candidate | 21 | 0.999999 | -0.074597 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| GPS2 | 0.0233022 | C_exploratory_kd_candidate | 22 | 0.931377 | 0.21261 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| ZNF706 | -0.0451039 | C_exploratory_kd_candidate | 23 | 0.999999 | -0.18045 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| MPG | -0.0355505 | C_exploratory_kd_candidate | 25 | 0.999999 | -0.09514 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| NEMF | -0.0669372 | C_exploratory_kd_candidate | 26 | 0.151039 | 0.30293 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| EPB41L2 | -0.0100547 | C_exploratory_kd_candidate | 27 | 0.999999 | -0.075102 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| NBEAL1 | -0.0192786 | C_exploratory_kd_candidate | 28 | 0.999999 | 0.02296 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| ARID5B | -0.0265471 | C_exploratory_kd_candidate | 29 | 0.999999 | 0.0063724 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| PSMD9 | 0.0481448 | C_exploratory_kd_candidate | 30 | 0.957081 | 0.08609 | False | NO_PUBLISHED_FUSION_HIT_AT_FDR_0.1 |
| TAF11 | 0.00154048 | C_exploratory_kd_candidate | 52 | 0.093536 | 0.41568 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |
| KEAP1 | -0.0299126 | C_exploratory_kd_candidate | 109 | 0.006876 | 0.98717 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |
| DRAP1 | -0.0838194 | C_exploratory_kd_candidate | 250 | 0.00055 | 2.5053 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |
| TADA3 | -0.016035 | C_exploratory_kd_candidate | 278 | 0.008168 | 1.6521 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |
| DCTN1 | -0.0557471 | C_exploratory_kd_candidate | 286 | 0.092251 | 0.71721 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |
| PNKP | -0.0388128 | C_exploratory_kd_candidate | 376 | 0.093954 | 1.1401 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |
| HSPA14 | -0.0343068 | C_exploratory_kd_candidate | 377 | 0.055718 | 1.1322 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |
| DPH1 | -0.113247 | C_exploratory_kd_candidate | 410 | 0.084047 | 0.63308 | True | KO_IMPAIRS_HUMAN_MYOBLAST_FUSION |

The feasible queue is significantly depleted, rather than enriched, for published fusion-impairing hits. This is a bounded safety-consistency result and may partly reflect the queue's existing dependency/essentiality filters; it is not evidence of DMD efficacy. Every remaining fusion-risk flag should be deprioritized or explicitly tested before wet-lab nomination.

## Processed-data feasibility boundary

The GEO cell-metadata file contains 2,782,263 raw barcode rows but does not expose the guide assignment claimed in the GEO description. The 3.7-GB Matrix Market file is published without a separate gene-annotation file. Reconstructing per-cell perturbation identities therefore requires the eight raw gRNA-library SRA runs (520.6 million spots; 157.2 Gbases) and the Parse Biosciences pipeline or clarification from the data generators. The published gene-level fusion phenotype table is complete enough for this bounded external stress test.

## Statistical fallacy scan (11/11 checked)

- Simpson's paradox: no donor-stratified reversal can be tested because the published gene-level screen summary is aggregated; this is a limitation, not evidence of homogeneity.
- Ecological fallacy: gene-level screen effects are not used to infer individual-cell or patient outcomes.
- Berkson's paradox: the screen universe was restricted to genes expressed in myoblasts; transport to untested genes is not claimed.
- Collider bias: the overlap is defined by assay availability, and no post-outcome covariate is adjusted for.
- Base-rate neglect: the 250/7,197 published hit rate and overlap-specific hit count are reported with enrichment tests.
- Regression to the mean: there is no pre/post extreme-score enrolment.
- Survivorship bias: the split-toxin assay enriches surviving fusion-defective cells by design; the endpoint is interpreted as fusion impairment, not general benefit.
- Look-elsewhere effect: four prespecified tests are all reported and Holm-corrected.
- Garden of forking paths: FDR < 0.1 follows the publication's declared hit threshold; top rescue decile is a declared sensitivity threshold but remains exploratory.
- Correlation is not causation: concordance or enrichment is not interpreted causally.
- Reverse causality: no directional causal claim links HepG2 rescue score and myoblast fusion phenotype.