Integrating Single-Sample Networks and Genetic Algorithm to Estimate Radiation Dose in Multiple Tissues of Mice under Space Environment.

Exposure to space radiation poses various health risks, so accurately estimating radiation dose in space is crucial. Herein, we integrated 301 transcriptomic profiles from 30 spaceflight datasets and developed radiation-dose estimation models for the space environment using a genetic algorithm. Two models were constructed in this work: one using gene expression fold changes as input (fold change model) and the other using gene degrees as input (degree model). Of note, we initially constructed a single sample network (SSN) for each spaceflight sample, respectively, and the degrees that represented the node (gene) features were extracted from the SSNs. Moreover, we not only constructed estimation models applicable to all tissues (overall models) but also developed specific models for each tissue (tissue models), enabling our models to be used across various task scenarios. According to the experimental results, all models demonstrate excellent performance in radiation dose estimation during spaceflight, and our genetic algorithm models achieve good predictive performance with a limited number of genes. We identified radiation-responsive genes, mainly involved in DNA repair, cell cycle, protein/amino acid metabolic pathways, energy metabolic pathways, nervous system development and differentiation, and cancer pathways. Through the expression and interaction patterns of these genes, we found that the space radiation environment could induce health risks such as cancers, psychiatric/neurological disorders, liver injury/toxicity disorders. In summary, the presented approach yields promising results for estimating radiation doses and supports the assessment of radiation risks in space environments.