Changes Using Machine Learning on Nasa Multi-Omics Data Publisher



Ansari M ; Hajihashemi A ; Rafienia M
Source: 32nd National and 10th International Iranian Conference on Biomedical Engineering, ICBME 2025 Published:2025

Abstract

Spaceflight imposes distinct physiological and environmental stresses that disrupt the human skin microbiome a vital interface mediating barrier integrity, immune signaling, and host adaptation. This study introduces a novel integrative machine learning framework that systematically connects NASA multi-omics datasets (GLDS-566 and OSD-574, Inspiration4 mission) to predict spaceflight-associated microbial shifts at a systems level. Using four pre-flight and four post-flight astronaut skin swab samples, microbial abundance profiles were modeled through a hybrid feature selection strategy combining mutual information and recursive feature elimination. Among four tested classifiers Logistic Regression, Random Forest, Support Vector Machine, and Naive Bayes Logistic Regression achieved the highest performance (accuracy =0.875, quadmF 1=0.865) quad under repeated cross-validation, emphasizing the capability of simpler models to capture subtle microbiological patterns within small, high-dimensional datasets. The framework identified four discriminative Taxonomic IDs (1283, 1654, 2642494, 543736), primarily linked to Staphylococcus and Corynebacterium, which are central to maintaining skin homeostasis. Their differential abundance between pre-and post-flight samples suggests microbial adaptations to spaceflight stressors such as microgravity, radiation, and humidity fluctuations mechanisms analogously observed in biomaterial surface adaptation systems. Unlike prior studies focusing solely on descriptive microbial variation, the present work advances a predictive modeling approach, enabling microbiome-based biomarker discovery for precision astronaut health monitoring. This proof-of-concept demonstrates the feasibility of autonomous microbial surveillance in space environments and establishes a reproducible computational foundation for future integration with broader datasets like the Space Omics and Medical Atlas (SOMA) to enhance long-duration mission healthcare resilience. © 2025 IEEE.