Predicting Bone Metastasis Using Gene Expression-Based Machine Learning Models

Area of Science:

• Computational biology
• Bioinformatics
• Machine Learning in Oncology

Background:

• Bone metastasis (BM) is common in breast and prostate cancers, causing severe pain and reduced quality of life.
• Current treatments focus on managing symptoms, but predicting BM risk for early intervention remains a challenge.

Purpose of the Study:

• To develop a computational pipeline integrating machine learning (ML) and deep learning (DL) for predicting bone metastasis development.
• To identify key genetic predictors for early risk assessment and intervention strategies.

Main Methods:

• A network-based computational pipeline incorporating ML/DL was developed.
• Several machine learning models were constructed, with a deep neural network (DNN) showing the highest performance.
• Model robustness was validated using an external TCGA dataset.

Main Results:

• The DNN model achieved a high prediction accuracy (AUC of 92.11%) using 34 key genes.
• External validation demonstrated robust performance with an AUC of 85.78%.
• The identified genes showed potential BM-related functionality, offering insights into metastasis drivers.

Conclusions:

• The developed AI tool shows promise for predicting bone metastasis across different primary cancer sites.
• The findings highlight the potential of computational approaches for early BM detection and understanding underlying mechanisms.