Related Experiment Video
Updated: Jun 24, 2025
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Mining bone metastasis related key genes of prostate cancer from the STING pathway based on machine learning
Guiqiang Li1,2, Runhan Zhao3, Zhou Xie3
1Department of Urology, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Background:
Prostate cancer (PCa) is the second most prevalent malignant tumor in male, and bone metastasis occurs in about 70% of patients with advanced disease. The STING pathway, an innate immune signaling mechanism, has been shown to play a key role in tumorigenesis, metastasis, and cancerous bone pain. Hence, exploring regulatory mechanism of STING in PCa bone metastasis will bring novel opportunities for treating PCa bone metastasis.
Methods:
First, key genes were screened from STING-related genes (SRGs) based on random forest algorithm and their predictive performance was evaluated. Subsequently, a comprehensive analysis of key genes was performed to explore their roles in prostate carcinogenesis, metastasis and tumor immunity. Next, cellular experiments were performed to verify the role of RELA in proliferation and migration in PCa cells, meanwhile, based on immunohistochemistry, we verified the difference of RELA expression between PCa primary foci and bone metastasis. Finally, based on the key genes to construct an accurate and reliable nomogram, and mined targeting drugs of key genes.
Results:
In this study, three key genes for bone metastasis were mined from SRGs based on the random forest algorithm. Evaluation analysis showed that the key genes had excellent prediction performance, and it also showed that the key genes played a key role in carcinogenesis, metastasis and tumor immunity in PCa by comprehensive analysis. In addition, cellular experiments and immunohistochemistry confirmed that overexpression of RELA significantly inhibited the proliferation and migration of PCa cells, and RELA was significantly low-expression in bone metastasis. Finally, the constructed nomogram showed excellent predictive performance in Receiver Operating Characteristic (ROC, AUC = 0.99) curve, calibration curve, and Decision Curve Analysis (DCA) curve; and the targeted drugs showed good molecular docking effects.
Conclusion:
In sum, this study not only provides a new theoretical basis for the mechanism of PCa bone metastasis, but also provides novel therapeutic targets and novel diagnostic tools for advanced PCa treatment.
