Methodology Showcase Series: NASA Artemis II Human Research Data Methodology Challenge
Four astronauts. Three studies. One dataset that breaks most conventional statistical approaches.
The NASA Artemis II Human Research Data Methodology Challenge asked participants to solve a problem unique to deep space research: how do you extract meaningful scientific insight from a sample size of four, spanning multiple physiological systems, data modalities, and time points?
This series brings the winning approaches to life across four video episodes, each one grouped around a shared methodological theme. Watch the episodes below, and read a short summary from each winning participant.
The video recordings are provided by the challenge participants and describe their approaches in their own words.
Episode #1: Small-Sample Inference & External-Data Borrowing
Featured in this video:
Praniil Nagaraj | 1st place
Transfer-Calibrated Normative ModelingAffiliation: Purdue University Project Approach: "This project treats the Artemis II challenge - four astronauts measured across hundreds of features over time - as a high-resolution case series rather than a failed population study, applying a stack of proven small-sample, high-dimensional methods (variance-borrowing empirical-Bayes tests, exact permutation inference, shrinkage-based multivariate scoring, and Gaussian-process recovery modeling) on a common calibrated scale. We demonstrate it on real four-person spaceflight data from the Inspiration4 crew, recovering a coherent acute inflammatory signature and per-crew recovery trajectories - with automated quality control that catches assay drift rather than misreading it as biology."
OJ Ochiai | 4th place
Bayesian N-of-1 with Dynamic BorrowingAffiliation: Independent Researcher Project Approach: "This project applies Bayesian hierarchical N-of-1 modeling with dynamic borrowing to the Artemis II problem of extracting rigorous inference from only four astronauts across many measurements. Applied unchanged to the Inspiration4 mission (the only existing real n=4 spaceflight cohort), the model independently recovers textbook spaceflight physiology from serum and urinary biomarkers — demonstrating that a single small-sample framework can honestly quantify what four astronauts' worth of high-dimensional data can, and cannot, support."
Ismael AL-Hadhrami | 5th place
Dynamic Empirical-Bayes CounterfactualAffiliation: University of Virginia Project Approach: "I developed a Dynamic Empirical-Bayes Bayesian Counterfactual Simulation Framework to analyze small-sample, high-dimensional astronaut health data from NASA's Spaceflight Standard Measures program. The methodology leverages proxy cohorts, empirical priors, Bayesian borrowing, and robust longitudinal analysis to identify meaningful physiological changes while overcoming small sample sizes. This framework provides a scalable and interpretable approach for supporting evidence-based decisions in NASA's Human Research Program."
Episode #2: Multi-Modal Fusion & Latent-Factor Integration
Featured in this video:
Estrella (Star) Perez | 2nd place
STELLAR (Systematic Trajectory Estimation via Layered Longitudinal Analysis and Regression) Affiliation: Independent Researcher Project Approach: "STELLAR is a three-layer Bayesian pipeline designed to extract meaningful signal from the kind of data NASA's Human Research Program collects on missions like Artemis II: small crews, high-dimensional measurements, multiple biological systems tracked together over time. It combines multi-omics factor analysis v2, Bayesian hierarchical modeling, and Gaussian Process regression, with two purpose-built diagnostic metrics: the Physiological Data Diversity Index (PDDI) and Physiological Complexity Index (PCI), to characterize dataset diversity and physiological complexity. The methodology is built specifically for the small-sample reality of human spaceflight research, where traditional statistical approaches often fall short."
Waste Parrot Technologies | 3rd place (Naseem Khan)
Unified Hierarchical Bayesian SpineAffiliation: Waste Parrot Technologies Project Approach: "Waste Parrot Technologies developed a unified hierarchical Bayesian framework for extracting reliable insight from small-sample, high-dimensional spaceflight health data. The approach condenses large numbers of correlated biomarkers into a small set of interpretable signals and models each astronaut individually rather than averaging across the crew, recovering patterns that conventional methods miss with only four subjects. We demonstrated it end-to-end on public proxy data for two Artemis II studies, Immune Biomarkers and ARCHeR, producing calibrated early-warning flags for unusual physiological readings, designed for real mission constraints such as far-side communication blackout."
Ashlei Lewis | 9th place
Integrated Multi-Study Health Profiling Affiliation: Neuroviu Project Approach: "This project uses a hierarchical Bayesian multi-block factor analysis to model each Artemis II astronaut individually while leveraging reference data from ISS, Inspiration4, and terrestrial analog missions to improve statistical reliability. The approach is designed for extremely small, high-dimensional datasets, enabling personalized mission-stress profiling, uncertainty estimation, and detection of integrated physiological and behavioral changes without relying on traditional group-based statistical tests. Validation on the PhysioNet MMASH dataset demonstrated a reproducible end-to-end pipeline suitable for deep-space human research."
Episode #3: Individualized Trajectory, Mechanistic & ML Modeling
Featured in this video:
Aaron Krasinski | 7th place
Mechanism-Constrained AI for Vision Risk (SANS)Affiliation: Massachusetts Institute of Technology Project Approach: "I designed a multi-task deep learning model that simultaneously predicts SANS from longitudinal bed rest MRI data and regresses six quantitative biomarkers characterizing glymphatic dysfunction. This kind of approach could let researchers take tools traditionally suited to large-sample data and use them to extract meaningful, biologically grounded insight into the pathology of conditions that affect only a small minority of the population, such as SANS."
Episode #4: Operational Monitoring, Behavioral Health & Decision Support
Featured in this video:
Ivan Lavlinski | 6th place
Bayesian State-Space Decision Layer Affiliation: Independent Researcher Project Approach: "My project is a hierarchical Bayesian framework for tracking the health of the Artemis II crew, built for a hard case: only four astronauts, each measured across dozens of physiological systems, the exact setting where standard statistics break down. It combines 43 established methods, organized in four layers and drawing on twelve public datasets, into a single system a flight surgeon can act on. I tested the whole thing on real spaceflight data from the Inspiration4 mission, and it runs the same way whether the mission lasts ten days or the several months it takes to reach Mars."
Carm Hermosilla | 10th place
Akasi (Interactive Data Interface)
Affiliation: University of California, Irvine
Project Approach: "Akasi is a visual health data dashboard that showcases a variety of data analysis methods and models to better understand trends between sleep & actigraphy at a 6-degree tilt as well as fluctuations of macronutrient levels over time. This is intended to be a visual aid to modeling & analyzing, with interactive graphs via the Streamlit web app service. With preprocessing tools & a solid framework, further physiological data can be incorporated smoothly into Akasi, with the intention to model & analyze trends between different datasets."
Vision One | 8th place (Sushanta Khadka, Arpan Bom, Pranisha Karki)
Integrated Health Monitoring Affiliation: Vision One Project Approach: "Vision One developed an Integrated Bayesian Multi-Modal Longitudinal Physiological Analysis Framework to help NASA extract meaningful insights from the Artemis II human research dataset despite having only four astronauts. By combining physiological, cognitive, immune, sleep, activity, imaging, and environmental data into a unified analytical framework, our approach enables personalized health monitoring, early detection of fatigue and stress, and supports future Artemis, Gateway, lunar surface, and Mars exploration missions."
Learn more about the NASA Artemis II Human Research Data Methodology Challenge at hrpdatachallenge.com.
