Datasets:

jang1563
/

genelab-benchmark

Tissue	Best AUROC	Method	Feature	perm_p	Significant
Thymus	0.948	PCA-LR	KEGG	<0.05	Yes*
Colon	0.921	PCA-LR	KEGG	<0.05	Yes*
Lung	0.901	PCA-LR	Gene	<0.05	Yes*
Kidney	0.829	ElasticNet-LR	Hallmark	<0.01	Yes**
Eye	0.823	PCA-LR	Hallmark	—	—
Skin	0.819	PCA-LR	Gene	—	—
Gastrocnemius	0.776	PCA-LR	Gene	—	—
Liver	0.670	PCA-LR	Gene	—	—

Classifier Rankings (Gene-level mean across 8 tissues)

Rank	Classifier	Gene Mean AUROC
1	PCA-LR	0.776
2	ElasticNet-LR	0.762
3	LightGBM	~0.72
4	XGBoost	~0.71
5	Random Forest	~0.70
6	SVM-Linear	~0.69
7	TabNet	0.527
8	SVM-RBF	0.510

Key v4 Findings

40/256 evaluations significant at p<0.05; **6/8** tissues have >=1 significant result
PCA-LR best overall; deep learning (TabNet) and kernel methods (SVM-RBF) worst
Pathway features improve: kidney (0.584->0.829), thymus (0.908->0.948), eye (0.697->0.823)
Gene features better for: skin (0.819), lung (0.901)
v4 expanded controls: BC/VC included (liver 261 vs v1's 193 samples)
v1 PCA-LR liver AUROC reproduced exactly (0.5870 vs 0.5871) using task folds

v4 Label Encoding

Flight/FLT -> 1
GC/Ground Control/Ground/Basal/BC/VC/Vivarium -> 0
AG (Artificial Gravity) -> excluded

v1 Results (6 tissues, original analysis)

Hypothesis Results

Hypothesis	Statement	Verdict	Key Evidence
H1	Liver has the most consistent cross-mission spaceflight transcriptome	REFUTED	Thymus(0.860) >> Liver(0.577). Thymus and gastrocnemius Tier 1.
H2	Cross-mission transfer failure from biological diversity, not batch	SUPPORTED	NES conservation correlates with transfer AUROC (r=0.9 for 5 tissues excl. gastrocnemius; original 4-tissue r=1.0). D3 pathway F1=0.06 (batch-invariant).
H3	Pathway-level preserves spaceflight response better than gene-level	CONDITIONALLY SUPPORTED	Kidney rescue (0.43→0.74), Eye (0.79→0.92). But liver→thymus anti-predicts (AUROC<0.5). Tissue-pair dependent.

Category B: Cross-Mission Transfer (PCA-LR, AUROC)

Tissue	Mean AUROC	95% CI	N Missions	N Pairs	Tier
Thymus	0.860	[0.763, 0.953]	4	12	1
Gastrocnemius	0.801	[0.653, 0.944]	3	6	1
Skin	0.772	[0.691, 0.834]	3	6	2
Eye	0.754	[0.688, 0.838]	3	6	2
Liver	0.577	[0.492, 0.666]	6	30	3
Kidney	0.555	[0.397, 0.681]	3	6	3

Thymus vs Liver Δ = 0.283. Permutation tests: Thymus vs Liver p=0.001, Gastro vs Liver p=0.048, Skin vs Liver p=0.032.

Category A Detection Significance (BH-FDR corrected)

Tissue	AUROC	Raw p	FDR q	Significant?
Skin	0.821	0.002	0.012	Yes
Gastrocnemius	0.824	0.026	0.074	No
Thymus	0.923	0.037	0.074	No
Eye	0.789	0.063	0.095	No
Liver	0.670	0.091	0.109	No
Kidney	0.432	0.281	0.281	No

Note: Only skin survives BH-FDR correction at α=0.05. However, all top-4 tissues have AUROC > 0.7 (GO threshold). High AUROC with modest significance reflects small fold counts (3-4 folds per tissue), not weak signal.

Category C: Cross-Tissue Transfer (3 Methods, AUROC)

Pair	Method A (Gene)	Method B (DEG)	Method C (Pathway)	Best
C1 liver→kidney	0.730 [0.62, 0.83]	0.441 NS	0.483 NS	A
C2 liver→gastro	0.563 NS	0.676*	0.867 [0.72, 0.98]	C
C3 liver→thymus	0.350 NS	0.621*	0.184 (anti)	B
C4 thymus→kidney	0.585 NS	0.539 NS	0.690 [0.58, 0.79]	C

H3 test: Method C wins 2/4 pairs (C2, C4). C vs A mean diff = -0.001 (essentially tied overall).

Category D: Condition/Confounder Prediction (macro-F1)

Task	Tissue	N	Gene F1	Gene p	Pathway F1	Pathway p	Interpretation
D3 Mission ID (6-class)	Liver	264	1.000 [1.00, 1.00]	<0.001	0.056 [0.04, 0.07] NS	1.0	Perfect batch separation (gene); batch-invariant (pathway)
D4 Strain (2-class)	Thymus	34	0.892 [0.48, 1.00]	0.004	0.817 [0.47, 1.00]	0.015	Strain detectable even from GC-only samples. EXPLORATORY (n_minority=3)
D5 Hardware RR vs MHU	Liver	264	1.000 [1.00, 1.00]	<0.001	0.386 [0.36, 0.41] NS	1.0	Perfect gene separation; collinear with D3 (hardware derived from mission)
D5 Hardware RR vs MHU	Thymus	92	1.000 [1.00, 1.00]	<0.001	0.352 [0.31, 0.39] NS	1.0	Perfect gene separation; collinear with D3
D6 Gravity (3-class)	Liver	9	0.886	0.002	0.413 NS	0.354	uG separable from gene expression
D6 Gravity (3-class)	Thymus	9	0.657	0.037	0.641 (p=0.052)	0.052	Gene ≈ Pathway for gravity detection

Confounder Hierarchy

D3 (mission F1=1.0) >= D5 (hardware F1=1.0, collinear) >= D4 (strain F1=0.89, exploratory n=3)
All pathway F1 ≈ 0.05-0.41 → pathways resist confounder detection (batch-invariant)

Key insight: D5 hardware prediction is perfect but collinear with D3 — hardware type (RR vs MHU) is a deterministic function of mission ID. D5 F1 should be interpreted as an upper bound of D3, not independent evidence. D4 strain effect is detectable but exploratory (minority class n=3).

J5: Gene-level vs Pathway-level Comparison (12 tasks)

Category A — Spaceflight Detection (LOMO, AUROC)

Tissue	Gene	Pathway	Diff (P-G)	Winner
Liver	0.670	0.574	-0.096	Gene
Gastrocnemius	0.824	0.688	-0.137	Gene
Kidney	0.432	0.743	+0.311	Pathway
Thymus	0.923	0.879	-0.044	Gene
Eye	0.789	0.915	+0.125	Pathway

Mean diff (Cat A): +0.032 (essentially tied)

Across All Categories

Category	N	Gene wins	Pathway wins	Mean diff
A (Detection)	5	3	2	+0.032
C (Cross-tissue)	4	2	2	-0.001
D (Condition, D3+D6 original)	3	3	0	-0.478
D (Condition, full D3-D6)	6	6	0	-0.462
Total (original 12)	12	8	4	-0.106
Total (expanded 15)	15	11	4	-0.174

Note: D4/D5 all show gene >> pathway, consistent with D3/D6 pattern. Pathways systematically resist confounder/batch detection.

NES Conservation vs Cross-Mission Transfer

Tissue	NES Mean r	Transfer AUROC	N fGSEA missions
Thymus	0.619	0.860	3
Eye	0.335	0.754	3
Skin	0.147	0.772	3
Liver	0.059	0.577	6
Gastrocnemius	0.057	0.801*	2
Kidney	0.048	0.555	3

*Gastrocnemius outlier: only 2/3 missions have fGSEA data (RR-5 no DGE). †Skin: RR-7 DGE absent; fGSEA on RR-6, MHU-2_dorsal (GLDS-238), MHU-2_femoral (GLDS-239) only. Excluding gastrocnemius: rank-order correlation for 5 tissues (thymus/eye/skin/liver/kidney) Spearman r = 0.9 (skin NES rank 3rd vs transfer rank 2nd — partial outlier). Original 4-tissue finding (thymus/eye/liver/kidney, excl gastrocnemius) maintains perfect rank concordance (Spearman r = 1.0).

Biological Validation (fGSEA Hallmark, all tissues PASS)

Tissue	Top Differentially Enriched Pathways (FLT vs GC)	Consistency
Liver	OXIDATIVE_PHOSPHORYLATION, FATTY_ACID_METABOLISM	Literature-concordant (direction varies by mission)
Thymus	E2F_TARGETS, G2M_CHECKPOINT, IFN-gamma	Thymocyte proliferation
Gastrocnemius	OXIDATIVE_PHOSPHORYLATION, MYOGENESIS	Muscle metabolism (direction varies by mission)
Kidney	MTORC1_SIGNALING, CHOLESTEROL_HOMEOSTASIS	Renal metabolism
Eye	OXIDATIVE_PHOSPHORYLATION (dominant 3/3 missions)	Retina metabolic demand
Skin	E2F_TARGETS, G2M_CHECKPOINT, EPITHELIAL_MESENCHYMAL_TRANSITION	Cell proliferation + ECM remodeling (2/3 missions consistent)

Note: "Top Differentially Enriched" = highest |NES| across missions. Enrichment direction (up/down in spaceflight) may vary by mission for liver and gastrocnemius due to mission-specific biological variability. See individual fGSEA result files in processed/fgsea/ for per-mission NES values and directions.

Tier 2: Geneformer (Mouse-GF) vs Classical Baseline

Mouse-Geneformer (6L BERT, 56K mouse gene vocab, pretrained on 30M scRNA-seq cells) fine-tuned on bulk RNA-seq LOMO folds (10 epochs, batch=16, lr=2e-5, freeze=4/6 layers).

Task	Tissue	Geneformer AUROC	Baseline AUROC	Baseline Model	Delta	Winner
A1	Liver	0.486 ± 0.074	0.588	LR ElasticNet	-0.102	Baseline
A2	Gastrocnemius	0.382 ± 0.054	0.907	LR ElasticNet	-0.525	Baseline
A3	Kidney	0.452 ± 0.080	0.521	LR ElasticNet	-0.069	Baseline
A4	Thymus	0.495 ± 0.233	0.923	PCA-50 + LogReg	-0.428	Baseline
A5	Skin	0.557 ± 0.087	0.821	LR ElasticNet	-0.265	Baseline
A6	Eye	0.484 ± 0.117	0.789	PCA-50 + LogReg	-0.305	Baseline
Mean	6 tissues	0.476	0.758	—	-0.283	Baseline

Interpretation: Classical ML wins 6/6 tissues (sign test p=0.016). Geneformer performs near chance level (0.5) on small-n bulk RNA-seq (train n=30-100). This is consistent with literature — foundation models pretrained on single-cell data do not automatically transfer to small-sample bulk transcriptomics tasks.

Note: Table shows best baseline per tissue for fair comparison. Publication figures use unified PCA-LR baseline (mean 0.743) for cross-figure consistency with Category A/B results.

Tier 2: scGPT (whole_human) vs Classical Baseline

scGPT-whole_human (12L Transformer, 512d hidden, 8 heads, pretrained on 33M human CellXGene cells) fine-tuned on mouse bulk RNA-seq LOMO folds via ENSMUSG→human gene symbol ortholog mapping. Training: 10 epochs, batch=8, lr=1e-4, freeze=10/12 layers (flash_attn disabled for PyTorch 2.1 compatibility).

Note on reliability: Folds with n_test ≤ 8 (MHU-1 thymus, RR-9 gastro) produce highly variable AUROC estimates and should be interpreted with caution. Large-n folds (RR-8 liver n=103, RR-7 kidney n=94, RR-7 skin n=30) are most reliable.

Task	Tissue	scGPT AUROC	Geneformer AUROC	Baseline AUROC	Δ vs GF	Δ vs Baseline	Winner
A1	Liver	0.628 ± 0.283	0.486	0.588	+0.142	+0.040	scGPT
A2	Gastrocnemius	0.685 ± 0.305	0.432	0.801	+0.253	-0.116	Baseline
A3	Kidney	0.556 ± 0.195	0.432	0.538	+0.124	+0.018	scGPT
A4	Thymus	0.782 ± 0.172	0.476	0.923	+0.306	-0.141	Baseline
A5	Skin	0.691 ± 0.050	0.532	0.821	+0.159	-0.130	Baseline
A6	Eye	0.650 ± 0.141	0.478	0.789	+0.172	-0.139	Baseline
Mean	6 tissues	0.666	0.476	0.758	+0.190	-0.092	Baseline

Interpretation: Classical ML wins 5/6 tissues vs scGPT (sign test p=0.109, ns). scGPT outperforms Geneformer by +0.190 AUROC across all tissues, suggesting human-pretrained 12L transformer captures more transferable features than mouse-specific 6L BERT. However, both FMs remain below classical ML baseline (scGPT: -0.092, Geneformer: -0.283), confirming that pretrained single-cell FMs do not transfer reliably to small-n bulk transcriptomics. The performance gap narrows but does not close: Classical ML 6/6 > both FMs.

Key observation: scGPT shows higher variance (std=0.05–0.31) than Geneformer (std=0.05–0.23), partly reflecting ortholog mapping noise from human pretraining. Large-n reliable folds (liver RR-8 n=103: 0.468; kidney RR-7 n=94: 0.557; skin n=30–39: 0.636–0.737) suggest scGPT hovers near chance (0.5) on the most statistically robust estimates.

Results file: evaluation/scgpt_whole_human_all_tissues_summary.json

Held-Out Evaluation: A4 Thymus (OSD-515 / RR-23)

Reserved held-out test set for external benchmark evaluation. Train on 4 missions (MHU-1, MHU-2, RR-6, RR-9; n=67), test on RR-23 (n=16: 7 Flight, 9 GC). 27,541 common genes.

Model	AUROC	95% CI	p-value
LR ElasticNet	0.905	[0.672, 1.000]	0.005
Random Forest	0.905	[0.672, 1.000]	0.007
PCA-50 + LogReg	0.873	[0.609, 1.000]	0.011
Geneformer (Mouse-GF)	0.556	[0.265, 0.850]	—

Interpretation: Classical baselines achieve strong held-out performance (AUROC ~0.90, p<0.01), confirming thymus cross-mission generalization beyond LOMO. Geneformer remains near chance on held-out data, consistent with LOMO results (0.495). The held-out confirms thymus as the most robust tissue for spaceflight detection.

Tier 3: LLM Zero-Shot Classification

Three LLMs tested on zero-shot text-based spaceflight detection (no training, gene expression → text prompt → binary prediction).

Model	A1 Liver	A2 Gastro	A3 Kidney	A4 Thymus	A5 Skin	A6 Eye	Mean
PCA-LR (ref)	0.670	0.824	0.432	0.923	0.821	0.789	0.743
DeepSeek-V3	0.435	0.514	0.495	0.421	0.467	0.492	0.471
Gemini-2.5-Flash	0.523	0.438	0.494	0.602	0.580	0.393	0.505
Llama-3.3-70B	0.527	0.544	0.440	0.533	0.451	0.407	0.484

Interpretation: All 3 LLMs perform at chance level (mean 0.47–0.51). Text-based reasoning cannot replace numerical ML for transcriptomics classification. Protein-coding gene filter was applied to reduce prompt noise.

Multi-DB Pathway Comparison (LOMO, PCA-LR)

Tissue	Hallmark	KEGG	Reactome	MitoCarta	Best DB	Range
Thymus	0.879	0.899	0.922	0.846	Reactome	0.076
Gastro	0.688	0.713	0.755	0.627	Reactome	0.128
Skin	0.690	0.754	0.693	0.542	KEGG	0.212
Eye	0.915	0.625	0.658	0.478	Hallmark	0.437
Liver	0.574	0.639	0.614	0.555	KEGG	0.084
Kidney	0.743	0.665	0.779	0.641	Reactome	0.138

Key findings:

DB choice > model choice (AUROC range up to 0.437 for Eye)
No single DB dominates: Reactome best for 3 tissues, KEGG for 2, Hallmark for 1
MitoCarta consistently worst (specialized → low coverage)

Temporal & Biological Covariates

T1: ISS-T vs LAR Sacrifice Timing

Question: Can sacrifice timing (ISS-Terminal vs Live Animal Return) be detected from transcriptomics?

Confound warning (DD-18): ISS-T = RNAlater on-orbit, LAR = standard necropsy. Preservation method confounds biological timing.

Sub-task	Condition	Gene AUROC	Pathway AUROC	n
T1a RR-6 liver	FLT	1.000	0.960	20
	GC (baseline)	0.922	0.778	19
T1b RR-8 liver	FLT	0.930	0.920	35
	GC (baseline)	0.973	0.993	35
T1c RR-6 thymus	FLT	0.857	0.714 NS	17
	GC (baseline)	0.925	1.000	18

Verdict: GC AUROC ≥ FLT AUROC in most cases → ISS-T vs LAR difference dominated by preservation artifact, not biological timing. Cross-mission transfer (T1d) confirms: artifact is consistent across RR-6↔RR-8 (FLT gene AUROC 0.97–0.99, GC gene 0.84–0.96).

T2: LAR Recovery Signature

Mission	PCA Recovery R	Pathways Recovering	FLT_LAR flight_prob
RR-6	0.842 (partial)	12/26	0.185 (strong)
RR-8	0.652 (stronger)	25/27 (overshoot)	0.404 (moderate)

RR-8 shows strong recovery with overshoot past baseline (MYC targets V1 +2.49, Protein secretion +2.14). RR-6 shows immune rebound (IFN-α -2.36, Inflammatory -2.54).

T3: Age × Spaceflight Interaction (RR-8 Liver)

Comparison	Gene AUROC	Pathway AUROC	n
T3a: Overall OLD vs YNG	0.985	0.851	141
T3d: Spaceflight in OLD	0.945 [0.846, 1.00]	0.879	34
T3d: Spaceflight in YNG	0.679 [0.479, 0.86]	0.716	36
Delta (OLD - YNG)	+0.266	+0.163	—

Verdict: "Spaceflight amplifies aging" SUPPORTED (Δ=+0.266). T3c ANOVA: 0/50 significant Age×Spaceflight interactions at FDR<0.05 (underpowered, n=10/cell).

J2: DGE Pipeline Comparison

Question: Does the choice of DGE pipeline (DESeq2 vs edgeR vs limma-voom) affect downstream results?

Scope: 9 missions (6 liver + 3 thymus) × 3 pipelines = 27 DGE runs. Skin excluded (RR-7 has no raw counts).

Metric	Mean	Min	Max
Log2FC Spearman	0.926	0.790	1.000
Log2FC Pearson	0.895	0.706	1.000
DEG Jaccard (FDR<0.05)	0.600	0.000	1.000
GeneLab Replication	0.707	—	9 missions

Key findings:

Fold-change rankings are highly conserved across all three pipelines (Spearman 0.926)
DEG list overlap varies by pipeline stringency: limma-voom most liberal, edgeR most conservative
RR-3 liver: 0 DEGs across all pipelines (n=6+6, true biological null — GeneLab also found only 1 DEG)
RR-1 edgeR: 0 DEGs due to conservative multiple testing correction, but log2FC correlation >0.95 with DESeq2
GeneLab replication (our binary FLT-vs-GC vs GeneLab's multi-level contrasts): moderate agreement (Spearman 0.707) reflects different design matrices, not pipeline error

Verdict: Rankings consistent, DEG lists vary by stringency threshold. Pipeline choice has moderate impact on DEG calls but minimal impact on gene-level effect size rankings — consistent with J1 (pipeline version comparison).

Held-Out Evaluation: A5 Skin (OSD-254 / RR-7)

Second held-out test set. Train on 2 missions (RR-6, MHU-2; n=72), test on RR-7 (n=30: 10 Flight, 20 GC). 20,110 common genes. RR-7 is a 75-day mission (longest in skin dataset).

Model	AUROC	95% CI	p-value
LR ElasticNet	0.885	[0.745, 0.986]	<0.001
PCA-50 + LogReg	0.840	[0.679, 0.963]	0.001
Random Forest	0.777	[0.583, 0.929]	0.007

Cross-Tissue Held-Out Comparison:

Tissue	Mission	Duration	Best AUROC	n_test
Thymus	RR-23	30 days	0.905 (LR)	16
Skin	RR-7	75 days	0.885 (LR)	30

Interpretation: Skin held-out confirms strong generalization (AUROC 0.885, p<0.001), exceeding the LOMO mean (0.821). Both held-out tissues achieve AUROC > 0.85, validating cross-mission spaceflight detection beyond leave-one-out evaluation.

Pipeline Status

Component	Files	Status
fGSEA	80 (6 tissues × missions × 4 DBs incl. MitoCarta)	Complete
GSVA	88 (6 tissues × missions × 4 DBs, skin+thymus MHU-1)	Complete
Category A	6 tissues, PCA-LR LOMO	Complete
Category B	6 tissues, bootstrap CI + permutation	Complete
Category C	4 pairs × 3 methods	Complete
Category D	D3 + D4 + D5×2 + D6×2 (6 tasks)	Complete
J5	15 comparisons	Complete
NES Conservation	6 tissues × 4 DBs	Complete
Multi-DB LOMO	24 runs (6 tissues × 4 DBs)	Complete
NC1/NC2	Permutation + housekeeping controls	Complete
Cell 2020	5 tissues pathway validation	Complete
Geneformer	6 tissues, 22 LOMO folds (Mouse-GF)	Complete
scGPT	6 tissues, 21 LOMO folds (whole_human), mean AUROC=0.666	Complete
LLM Zero-Shot	3 providers × 6 tasks (18 evals)	Complete
Held-Out	A4 Thymus (RR-23) + A5 Skin (RR-7)	Complete
T1-T3 Temporal	ISS-T/LAR, Recovery, Age×Spaceflight	Complete
J2 DGE Pipeline	9 missions × 3 pipelines (DESeq2/edgeR/limma-voom)	Complete
v1 Figures	4 main + 4 supplementary (HTML/SVG)	Complete
v2 E1-E3	Cross-species NES, duration effect, cfRNA origin	Complete
v2 F1	I4 PBMC cell-type fGSEA (10 types × 50 pathways)	Complete
v2 Figures	3 integrated main figures (D3.js v7)	Complete
RRRM-1 scRNA	4 tissues (blood/eye/muscle/skin), 38K cells, annotated	Complete

v3 Results Summary

Foundation Model Comparison (7 tissues)

Model	Liver	Gastro	Kidney	Thymus	Eye	Lung	Colon	Mean
PCA-LR	0.670	0.824	0.432	0.923	0.789	—	—	0.758
scGPT	0.628	0.685	0.556	0.782	0.650	—	—	0.667
scFoundation	0.635**	0.691*	0.541	0.487	0.563	0.389	0.755	~0.58
UCE (seeded)	0.459	0.578	0.489	0.632*	0.550	0.555	0.449	~0.53
Geneformer	0.486	0.382	0.452	0.495	0.484	—	—	0.476

*p<0.05, **p<0.01. All FMs underperform PCA-LR baseline.

RRRM-2 scRNA-seq (F5)

Tissue	Best Cell Type	AUROC	Significance
PBMC	NK cell	0.845*	p<0.001
PBMC	T cell	0.752*	p<0.001
Spleen	B cell	0.562***	p<0.001
Bone marrow	All 14 types	0.27-0.54	No signal

Spatial Visium (F3, Brain OSD-352)

Negative result: Section AUROC=0.139, Animal AUROC=0.444
PC1 (42.5%) = slide batch effect, not spaceflight condition
Brain = genuine negative for spaceflight classification

Cross-Tissue Transfer (B_ext, 7x7 = 42 pairs)

Method A (gene) range: 0.35-0.80, liver->kidney best (0.73)
Method C (pathway) range: 0.43-0.87, liver->gastro best (0.87)

v4 Pipeline Status

Component	Status
Phase 1: 256 evaluations (8 tissues × 8 methods × 4 features)	Complete
Phase 2: Ablation studies (569 evals: feature count, PCA dims, sample size, bootstrap)	Complete
Phase 3: Friedman LOMO-6 meta-analysis (chi2=17.333, p=0.015)	Complete
Phase 4: SHAP multi-method interpretability	Complete
Phase 5: Python WGCNA (6 tissues), module preservation, STRING PPI enrichment	Complete
Phase 7: Publication figures (6 main + 5 supplementary HTML)	Complete
Phase 8: Manuscript preparation	In Progress

v5 Biological Interpretation

Immune Deconvolution (mMCP-counter, 8 tissues)

Tissue	Significant Cell Types (FDR<0.05)	Strongest Signal
Skin	6/14	Fibroblasts↑FLT, NK cells↑FLT
Kidney	2/14	—
Thymus	2/14	—
Liver, Gastro, Eye, Lung, Colon	0/14	No signal

Direction convention: positive Cliff's delta = higher in Flight vs. Ground.

Cross-Organ Signaling (OmniPath)

111 intercell-filtered ligand–receptor pairs (9 strict, 102 broad)
1 SHAP-active L–R pair identified
TF activity (CollecTRI + decoupler ULM): thymus 240 sig, skin 241, kidney 177, liver 105

Metabolic Flux (iMM1865 E-Flux + pFBA)

Tissue	FLT objective	GC objective	Difference
Thymus	15,695	14,696	999 (largest)
Liver	16,510	16,110	400
Gastrocnemius	—	—	varies
Kidney, Eye, Skin	—	—	varies

E-Flux normalized to [0,1] range; pFBA used to resolve LP degeneracy.

Drug Target Mapping (DGIdb v5 + ChEMBL)

834 WGCNA/SHAP consensus spaceflight genes → mouse→human ortholog mapping
271/834 (32.5%) human orthologs have known drug interactions (DGIdb)
1,284 FDA-approved drug–gene interactions (Tier 1)
200 investigational drug–gene interactions (Tier 3)
Thymus most druggable tissue (24.8%); 45 WGCNA modules enriched

Consensus Biomarker Panel (20 genes)

Scoring: SHAP rank (0–4) + WGCNA module membership (0–3) + multi-tissue (0–2) + druggability (0–1) + statistical significance (0–2)

Top Genes	Score	Notes
MUP22	5	Liver/skin WGCNA hub, SHAP top
Thrsp / THRSP	5	Metabolic hub
Apoa1	5	Liver SHAP + WGCNA
NPAS2	4	Circadian clock, gastro+skin modules
PER2	4	Circadian clock

Panel validation AUROC: gastro 0.806, liver 0.754, eye 0.728, colon 0.75, skin 0.70