Transcriptomic Signatures of Mechanotransduction in Osteogenesis and a Machine Learning-Based Predictive Model

Abstract

Traditional approaches in bone biology have relied on machine learning models that classify samples based on their overall osteogenic state, leading to limitations such as oversimplified sample labeling, data standardization biases, and batch effects. This study introduces a novel gene-centric classification framework that shifts the focus from predicting sample phenotypes to directly identifying genes involved in osteogenesis. By pre-labeling genes based on their known correlation with osteoporosis, we circumvent the need for large sample sizes and data normalization, effectively mitigating batch effects. Integrating bioinformatics and machine learning analyses of murine and human transcriptomic data, we developed a robust model that successfully predicted osteogenic differentiation-related genes. Validation using microgravity-responsive transcriptomic data led to the identification of six key hub genes (ETV4, ENPEP, ETS1, LRRFIP1, PLAUR, and PTX3) that are responsive to microgravity and promote osteogenic differentiation. This study provides valuable insights into the molecular mechanisms of osteogenesis and offers potential therapeutic targets for combating bone loss.