Study Methodology

Lettuce plants (Lactuca sativa cv. 'Outredgeous') were grown in Veggie pillows containing solid, porous, arcillite (calcined clay) substrate in two different blends (Blend 1: 100% sized to 600 µm–1 mm or Blend 2: a 1:1 ratio of that size to 1–2 mm) and controlled-release fertilizer (Nutricote 18-6-8, Florikan, Sarasota, FL, United States). Plants were grown for 33 to 56 days aboard the ISS under a red, blue, and green LED array providing approximately 200 µmol/m²/s of photosynthetically active radiation (PAR). Ground control experiments were conducted in controlled environment chambers at KSC under matching temperature, relative humidity, carbon dioxide concentration, and light regimes with a 24-72 hour delay.

Harvested leaves were stored at -80°C and returned to Earth. Nutritional assays included:

• Elemental Profiling: Inductively Coupled Plasma Optical Emission Spectroscopy (ICP-OES) for Fe, K, Na, P, S, Zn, Ca, Mg, Mn, and Cu (expressed as mg/g dry weight).
• Total Phenolic Content: Folin-Ciocalteu spectrophotometric assay, expressed as gallic acid equivalents (GAE mg/g dry weight).
• Anthocyanin Content: pH differential spectrophotometric assay, expressed as cyanidin-3-glucoside equivalents (mg/g dry weight).
• Oxygen Radical Absorbance Capacity (ORAC): Fluorescein decay antioxidant capacity assay against Trolox standard (µmol TE/g dry weight).

Analysis & ML Pipeline

Our analytical and machine learning pipeline is structured as follows:

1. Data Harmonization: Merged metadata (missions, conditions) and assay measurements, resolving column types and standardizing units.
2. Dimensionality Reduction: Standardized nutritional variables (z-score scaling) and applied Principal Component Analysis (PCA) to extract eigenvectors and coordinates, determining sample separation along the first two principal components.
3. Random Forest Classification: Trained an ensemble of 100 decision trees to classify Flight vs. Ground Control samples using the 11 features with complete coverage (nine elements plus phenolics and ORAC). Evaluated with Leave-One-Mission-Out Cross-Validation (LOMOCV), which tests generalization to an unseen spaceflight; accuracy was at chance. Feature importance is Gini (mean decrease in impurity); SHAP and permutation importance are computed only in the lettuce-only pipeline.
4. Tabular Foundation Model Benchmarking: Written for TabPFN (Tabular Prior-data Fitted Network) as described by Hollmann et al. (Nature 2025). Unlike gradient boosting trees or standard neural networks, TabPFN is a pre-trained transformer model optimized for zero-shot, prior-data fitted classification on small tabular datasets, It did not run here: without a license token the pipeline falls back to Gradient Boosting, and that fallback is the benchmark shown.
5. Antioxidant Capacity Prediction: Implemented Random Forest regression to predict biochemical outcomes (ORAC and Phenolics) based on multi-element profiles, measuring model performance via R² and root-mean-squared error (RMSE). Scored out-of-fold on lettuce (n = 24), R² was −0.90 for phenolics and −2.16 for ORAC, worse than predicting the mean.
6. Statistical Validation: Conducted non-parametric Mann-Whitney U tests for all analytes, applying false discovery rate (FDR, Benjamini-Hochberg) and Bonferroni corrections to prevent family-wise error rate inflation, and computed Cohen's d effect sizes.

Data Provenance (FAIR)

All data is sourced from NASA OSDR studies OSD-745 (Lettuce), OSD-655 (Mizuna Nutrition), and OSD-780 (Microbiology) following FAIR data principles.

Citations

% 1. Spaceflight Lettuce Nutrition & Safety (OSD-745)
@article{khodadad2020microbiological,
  title={Microbiological and nutritional analysis of lettuce crops grown on the International Space Station},
  author={Khodadad, Christina L and Hummerick, Mary E and Spencer, LaShelle E and Dixit, Anirudha R and Richards, Jeffrey T and Romeyn, Matthew W and Smith, Trent M and Wheeler, Raymond M and Massa, Gioia D},
  journal={Frontiers in Plant Science},
  volume={11},
  pages={199},
  year={2020},
  doi={10.3389/fpls.2020.00199}
}

% 2. Spaceflight Mizuna Pick-and-Eat Testing (OSD-655 & OSD-780)
@article{bunchek2024pick,
  title={Pick-and-eat space crop production flight testing on the International Space Station},
  author={Bunchek, Jess M and Hummerick, Mary E and Spencer, LaShelle E and Romeyn, Matthew W and Young, Millennia and Morrow, Robert C and Mitchell, Cary A and Douglas, Grace L and Wheeler, Raymond M and Massa, Gioia D},
  journal={Journal of Plant Interactions},
  volume={19},
  number={1},
  pages={2292220},
  year={2024},
  publisher={Taylor \& Francis},
  doi={10.1080/17429145.2023.2292220}
}