🌌 Executive Summary & Scientific Context
Future crewed lunar and Martian surface habitats, extravehicular activity (EVA) staging airlocks, and bioregenerative life-support systems (BLSS) will operate under NASA's reduced-pressure, oxygen-enriched Exploration Atmosphere: 56.54 kPa (8.2 psia) at 34% O₂ (normoxic oxygen partial pressure pO₂ = 19.22 kPa) and a staged pre-EVA setpoint of 70.33 kPa (10.2 psia) at 26.5% O₂ (pO₂ = 18.64 kPa).
The Fundamental Biophysical Asymmetry:
By the Chapman–Enskog kinetic theory of gases, binary molecular diffusivity scales inversely with total pressure (D ∝ 1/P_total).
Reducing cabin pressure to 56.54 kPa accelerates gas and water-vapor diffusion by 1.72× relative to ISS standard air (97 kPa), stripping the leaf boundary layer and driving acute evaporative demand.
While 34% O₂ eliminates crew hypoxia, plants still experience intense physical barosensory stress that oxygen supplementation alone cannot cure.
Here, we integrate the TabPFN v2 tabular foundation model (Hollmann et al., Nature 637, 319–326, 2025) across a 12-dataset multi-omics compendium spanning 281 microarray hybridizations, 40,860 root-tip single nuclei, 326 whole-plant gas-exchange records, and factorial pressure × humidity LC-MS metabolomics to decouple total barometric pressure from oxygen partial pressure.
🔬 Six-Workpackage Systems Architecture
Workpackage 1
Forensic Metadata Audit & Organ Classifier
Deduplicated 18 shared arrays between GSE87867 and GSE87903.
Coupled raw Affymetrix CEL headers with TabPFN v2 probabilistic auditing (AUC = 1.0000, Brier = 0.000002) to prove that 4 disputed 75 kPa samples (GSM2342806–809) are Shoots mislabeled as Roots in GEO, corrected 5 kPa root labels, and resolved replicate-A re-hybridizations across scanner batches (r = 0.9951–0.9964).
130 Unique Arrays • 22,746 Probesets • Table S1
Workpackage 2
TabPFN v2 Atmospheric Decoupling & Hub Networks
Benchmarked TabPFN v2 against LightGBM, Random Forest, ElasticNet, and deep MLP under condition-grouped cross-validation.
TabPFN v2 (PHE) achieved R² = 0.9134 on continuous hypobaria regression and ROC-AUC = 0.9623 on 4-class regime classification.
Orthogonal regression with limma moderation separated 1,003 Pure Hypoxia hubs from 380 Pure Barosensory hubs (CMI1, STZ, TCH4, LOX2, ACS6), 21 Additive Dual Hubs, and 12 Interaction Hubs.
R² = 0.9134 • 380 Barosensory Hubs • Tables S2A–S3
Workpackage 3
Cell-Type TRAP-seq & Single-Nucleus Mapping
Mapped atmospheric modules across 19 cell-type TRAP-seq translatomes (GSE14493/14502) and 40,860 root-tip single nuclei (GSE308672).
Demonstrated that hypoxia induces ERF-VII anaerobic modules across all 9 root lineages (+2.65 to +3.17 log2CPM), whereas barosensory wax, cuticle, and mechanosensing hubs localize specifically to pCER5 shoot epidermis (+2.21 and +1.21 log2).
19 Cell Types • 40,860 Single Nuclei • Tables S4A–B
Workpackage 4
Physiological & Metabolomic Phenotypic Bridge
Triangulated Ecotron whole-plant gas exchange (ZENODO16536514) at 62 kPa vs 101 kPa, revealing a +73.23% vapor pressure ratio surge driving a -76.25% collapse in stomatal conductance (g_sw) and a -55.93% drop in CO₂ assimilation.
Factorial pressure × humidity LC-MS metabolomics proved that 70% RH restores chloroplast oxylipin galactolipids (MGDG-OPDA) and reduces hypobaric state probability from 0.982 to 0.791.
326 Records • LC-MS Lipids/GSLs • Tables S5A–C
Workpackage 5
NASA Exploration Atmosphere Counterfactual Synthesis
Leave-One-Condition-Out (LOCO) validated TabPFN v2 predictions (mean r = 0.7387) for the NASA 56.54 kPa / 34% O₂ setpoint.
Counterfactual synthesis proves that 34% O₂ silences >88% of ERF-VII hypoxia induction, but physical barosensory stress (TCH4, LOX2, AOS) persists unabated—confirming that habitat humidity engineering (75–80% RH) is mandatory.
LOCO r = 0.7387 • Counterfactual Engine • Tables S6A–B
Workpackage 6
Plant PhysioSpace Stress Autodecoder Across Mission Profile
Projected the multi-omics compendium onto 10 canonical physiological stress dimensions across 6 spaceflight mission phases (Earth baseline, pre-EVA staging, nominal exploration habitat, decompression transients, emergency hypobaria, and BLiSS rescue).
Proved that nominal 56.5 kPa / 34% O₂ triggers isolated boundary-layer transpiration/osmotic stress (z = +1.86) rather than hypoxia (z = -0.16), and that 80% RH achieves total stress extinction (Δz = -1.90).
10 Stress Axes • 6 Mission Phases • Table S7
📊 Publication Figures (300 DPI High-Resolution)
Click on any figure thumbnail to view in full resolution.
Figure 1: Compendium Forensics & Audit
Harmonization of 130 unique hypobaric arrays across P_total vs pO₂ phase space, CEL audit of disputed 75 kPa samples, and TabPFN v2 organ classifier benchmark.
Figure 2: Physical Decoupling Benchmark
TabPFN v2 out-of-fold continuous regression benchmark (R² = 0.9134 on hypobaria, R² = 0.9733 on hypoxia), 4-class classification, and slice-based error analysis.
Figure 3: Molecular Hub Networks & SHAP
Orthogonal decomposition of 22,746 transcripts into 1,243 Hypoxia vs 380 Barosensory hubs, global TabPFN v2 SHAP attributions, and oxygen-rescue heatmap across 24 hubs.
Figure 4: Single-Nucleus & TRAP Spatial Atlas
TRAP-seq translatome enrichment across 19 cell types, 40,860 root-tip single-nucleus hypoxia trajectories, and metabolic sucrose rescue decoupling SnRK1 from ERF-VII.
Figure 5: Physiology & LC-MS Metabolomics
Ecotron stomatal gas exchange and chlorophyll fluorescence kinetics at 62 kPa, LC-MS galactolipid and glucosinolate profiling, and 70% RH metabolic rescue.
Figure 6: Counterfactual Synthesis & LOCO
Leave-One-Condition-Out validation across held-out regimes, TabPFN v2 counterfactual pathway trajectories under NASA setpoints, and hub gene barosensory persistence.
Figure 7: PhysioSpace Mission Profile Autodecoder
Continuous blue-white-red PhysioSpace stress autodecoder across 10 canonical stress axes and 6 astronautical mission phases, (Hypoxia vs Osmotic) 2D trajectory ordination, organ partitioning, and complete 80% RH boundary-layer stress rescue (Δz = -1.90).
📓 Executable Jupyter Notebooks
All five research workpackages are provided as fully executed, reproducible Jupyter notebooks containing code, documentation, statistical tests, and inline figures:
| Notebook | Workpackage Topic | Key Models & Analyses | Outputs |
|---|---|---|---|
| 01_data_harmonization.ipynb | Forensics & Deduplication | TabPFN v2 Organ Classifier, CEL Header Audit | Figure 1, Table S1 |
| 02_tabpfn_decoupling.ipynb | Atmospheric Decoupling & Hubs | TabPFN v2 (PHE) Regression, Multivariable Orthogonal Hubs, SHAP | Figures 2 & 3, Tables S2A–S3 |
| 03_cellular_mapping.ipynb | Cellular & Single-Nucleus Atlas | TRAP-seq Translatome, 40,860 Root Single Nuclei (GSE308672) | Figure 4, Tables S4A–B |
| 04_phenotypic_bridge.ipynb | Physiology & Metabolomics | Ecotron Gas Exchange, LC-MS Oxylipins, SnRK1 Proteomics | Figure 5, Tables S5A–C |
| 05_counterfactual_synthesis.ipynb | NASA Setpoints & PhysioSpace Stress Autodecoder | LOCO Validation, Counterfactual Simulator, 10-Axis PhysioSpace Autodecoder, 80% RH Rescue | Figures 6 & 7, Tables S6A–B, S7 |
📥 Supplementary Data Tables (CSV)
| Table ID | Description | Dimensions | Download |
|---|---|---|---|
| Table S1A | Full 148 GSM Forensic Audit Ledger & Affymetrix QC Status | 148 rows × 23 cols | CSV |
| Table S1B | Harmonized 130 Hypobaric Microarrays Forensic Audit | 130 rows × 28 cols | CSV |
| Table S2A | Continuous Physical Regression Benchmarks (95% Bootstrap CI) | 18 rows × 12 cols | CSV |
| Table S2B | 4-Class Atmospheric Regime Classification Benchmarks | 7 rows × 11 cols | CSV |
| Table S2C | Covariate Ablation Suite (Covariates vs Transcriptome vs Combined) | 6 rows × 7 cols | CSV |
| Table S2D | Slice-Based Error Analysis Across Biological Subgroups | 30 rows × 8 cols | CSV |
| Table S3 | Exploration Atmosphere Molecular Hub Nodes (Limma Moderated, 22,746 Probesets) | 22,746 rows × 18 cols | CSV |
| Table S4A | Cell-Type TRAP-seq and Single-Nucleus Lineage Atlas | 12 pathways × 24 cols | CSV |
| Table S4B | Top-80 Atmospheric Hubs Cell-Type & Single-Nucleus Localization | 80 rows × 26 cols | CSV |
| Table S5A | Zenodo 16536514 Physiological Gas Exchange Stats (62 kPa vs 101 kPa) | 12 traits × 11 cols | CSV |
| Table S5B | Zenodo 16536514 Pressure × Humidity Factorial Metabolomics | 14 compounds × 14 cols | CSV |
| Table S5C | PXD002353 SnRK1.1 Phosphoproteome & Transcriptome Integration | 5,454 proteins × 12 cols | CSV |
| Table S6A | NASA Exploration Atmosphere Counterfactual Predictions (with Conformal Bounds) | 6 setpoints × 44 cols | CSV |
| Table S6B | Leave-One-Condition-Out (LOCO) Model Generalization Benchmark | 20 rows × 8 cols | CSV |
| Table S7 | Plant PhysioSpace Stress Autodecoder Across Spaceflight Mission Profile | 12 rows × 19 cols | CSV |
📚 Citation, Authorship & Open Science
Authors: Richard J. Barker1,* and [Co-authors To Be Determined]2
Acknowledgements: The authors express sincere and deep gratitude to Dr. Anna-Lisa Paul and Dr. Robert J. Ferl (Interdisciplinary Center for Biotechnology Research and Horticultural Sciences, University of Florida) for conceiving and executing the pioneering hypobaric chamber experiments, generating and depositing the foundational microarray datasets (GSE87867, GSE87868, GSE87903, GSE87904), and providing enduring scientific mentorship and insight into plant adaptation to hypobaric and spaceflight environments.
If you use this compendium, TabPFN v2 model pipelines, or counterfactual predictions in your research, please cite:
@article{Barker_Exploration_Atmosphere_2026,
title = {Decoupling Total Barometric Pressure from Oxygen Partial Pressure in the Arabidopsis thaliana Spaceflight Exploration Atmosphere Using the TabPFN v2 Tabular Foundation Model},
author = {Barker, Richard J. and [Co-authors To Be Determined]},
journal = {npj Microgravity (Submitted)},
year = {2026},
url = {https://github.com/dr-richard-barker/arabidopsis-exploration-atmosphere-tabpfn}
}