Abstract Objective: This study aimed to identify key autophagy-related genes involved in osteoporosis using machine learning algorithms. Methods: Differential expression analysis between osteoporotic and control groups was performed using DESeq2. Candidate genes were obtained by intersecting the differentially expressed genes with a reference set of autophagy-related genes. Subsequently, LASSO regression, support vector machine, and random forest algorithms were applied to screen for characteristic genes. The overlapping genes resulting from all three methods were designated as feature genes. The in vitro expression of the feature genes was validated using RT-qPCR and western blot, and their functional roles in osteoblast differentiation and mineralization were assessed. Results and Discussion: Differentially expressed genes between the osteoporosis and control groups were identified based on adjusted p-values and |log2(fold change)|. Candidate genes were obtained by intersection with autophagy-related genes. Osteoporosis feature genes were screened from the candidate gene set using three well-recognized machine-learning algorithms. This approach yielded four feature genes: ADAM12, GZMB, SERPINF1, and TRIM6. In the osteoporosis cell model, ADAM12 and SERPINF1 were downregulated, whereas GZMB and TRIM6 were upregulated. ROC curve analysis indicated that all four feature genes exhibited strong discriminatory performance between osteoporotic and control samples (AUC >0.9), suggesting that the model possesses robust diagnostic ability. Functional experiments demonstrated that TRIM6 is involved in the regulation of osteoblast differentiation and mineralization. A gene regulatory network analysis revealed that EGR1, MAZ, and FOXA2 simultaneously regulate ADAM12, SERPINF1, and TRIM6. Drug sensitivity analysis indicated that increased TRIM6 expression correlates with sensitization to A-804598, istradefylline, SZ4TA2, sildenafil, MI-2, erismodegib, MK-0752, and necrostatin-7. Conclusions: Using machine learning algorithms, this study identified four osteoporosis-related autophagy-related feature genes: ADAM12, SERPINF1, GZMB, and TRIM6, which may serve as potential diagnostic markers. TRIM6 expression was found to be positively correlated with individual sensitivity to A-804598, istradefylline, SZ4TA2, and sildenafil, among others. Furthermore, TRIM6 was shown to participate in osteoblast differentiation and mineralization.

Integrative Machine Learning Screening and Functional Validation of Autophagy-Related Genes in Osteoporosis

期刊论文
发表日期：

2026年4月20日

查看全文

Jie Shi,
Yongxian Wan

摘要

Abstract

Objective: This study aimed to identify key autophagy-related genes involved in osteoporosis using machine learning algorithms. Methods: Differential expression analysis between osteoporotic and control groups was performed using DESeq2. Candidate genes were obtained by intersecting the differentially expressed genes with a reference set of autophagy-related genes. Subsequently, LASSO regression, support vector machine, and random forest algorithms were applied to screen for characteristic genes. The overlapping genes resulting from all three methods were designated as feature genes. The in vitro expression of the feature genes was validated using RT-qPCR and western blot, and their functional roles in osteoblast differentiation and mineralization were assessed. Results and Discussion: Differentially expressed genes between the osteoporosis and control groups were identified based on adjusted p-values and |log₂(fold change)|. Candidate genes were obtained by intersection with autophagy-related genes. Osteoporosis feature genes were screened from the candidate gene set using three well-recognized machine-learning algorithms. This approach yielded four feature genes: ADAM12, GZMB, SERPINF1, and TRIM6. In the osteoporosis cell model, ADAM12 and SERPINF1 were downregulated, whereas GZMB and TRIM6 were upregulated. ROC curve analysis indicated that all four feature genes exhibited strong discriminatory performance between osteoporotic and control samples (AUC >0.9), suggesting that the model possesses robust diagnostic ability. Functional experiments demonstrated that TRIM6 is involved in the regulation of osteoblast differentiation and mineralization. A gene regulatory network analysis revealed that EGR1, MAZ, and FOXA2 simultaneously regulate ADAM12, SERPINF1, and TRIM6. Drug sensitivity analysis indicated that increased TRIM6 expression correlates with sensitization to A-804598, istradefylline, SZ4TA2, sildenafil, MI-2, erismodegib, MK-0752, and necrostatin-7. Conclusions: Using machine learning algorithms, this study identified four osteoporosis-related autophagy-related feature genes: ADAM12, SERPINF1, GZMB, and TRIM6, which may serve as potential diagnostic markers. TRIM6 expression was found to be positively correlated with individual sensitivity to A-804598, istradefylline, SZ4TA2, and sildenafil, among others. Furthermore, TRIM6 was shown to participate in osteoblast differentiation and mineralization.