Harmonizing heterogeneous transcriptomics datasets for machine learning-based analysis to identify spaceflown murine...

Area of Science:

• Space biology
• Computational biology
• Genomics

Background:

• High-throughput molecular assays are used by NASA to detect physiological changes during spaceflight.
• Machine learning (ML) offers potential for analyzing complex, high-dimensional molecular data.
• Limited sample sizes in spaceflight studies hinder traditional ML applications.

Purpose of the Study:

• To develop and validate a method for integrating heterogeneous transcriptomics datasets from multiple spaceflight missions.
• To overcome sample size limitations for applying ML in spaceflight research.
• To accurately classify spaceflown versus ground control samples using ML on harmonized data.

Main Methods:

• Log transformation, scaling, and normalization of six heterogeneous mouse liver transcriptomics datasets (n=137).
• Application of ML methods to the integrated dataset to classify spaceflown vs. ground control animals.
• Comparison of ML-identified biological processes with classical omics analyses.

Main Results:

• ML models successfully classified spaceflown vs. ground control mice with an Area Under the Curve (AUC) of ≥0.87.
• The data harmonization approach effectively mitigated variability arising from different mission origins.
• Harmonized ML analysis revealed liver-specific biological processes consistent with study-by-study analyses.

Conclusions:

• Integrating heterogeneous, small-sample-size datasets is feasible for ML applications in spaceflight research.
• Harmonization techniques enable robust ML analysis despite inter-study variability.
• This approach enhances the utility of molecular data for understanding spaceflight's impact on physiology.