The Mitochondrial Signature for Predicting Outcome of Early-Stage Breast Cancer by Machine Learning.

Introduction: Breast cancer, a leading female cancer worldwide, can be influenced by mitochondrial dysfunction. Dysregulation of mitochondria by the nuclear genome may cause breast cancer initiation and progression. However, the comprehensive investigation of mitochondrial-related genes as prognostic marker for the overall survival of early-stage breast cancer patients is still limited.

Methods: To address this, we employed machine learning methods to identify a concise set of mitochondrial-related genes with high accuracy and reliability in predicting survival outcomes. Bulk transcriptome collected from Sweden Cancerome Analysis Network - Breast (SCANB) was divided into training and testing datasets and the Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) was included as the external validation cohort. The 1136 known mitochondrial-related genes were analysed using univariate Cox regression, bootstrap and Lasso Cox regression in the SCANB training cohort for model construction.

Results: We identified a 14-gene mitochondrial signature that independently predicts the survival outcome of breast cancer (adjusted hazard ratio [HR]: 2.08, 95% confidence interval [CI]: 1.20-3.62) in the SCANB dataset. A highly predictive nomogram was further constructed by integrating the mitochondrial signature with clinical variables, enabling robust prediction of overall survival at 1-, 3- and 5-year. This model demonstrated strong predictive capability in both the training cohort (the area under the receiver operating characteristic [ROC] curve [AUC]: 0.84, 0.79, 0.78) and validation cohort (AUC: 0.92, 0.83, 0.78).

Conclusion: In this study, we suggested a novel mitochondrial signature model by comprehensively analysing mitochondrial-related genes, which have the potential to accurately predict the clinical prognosis at the early stages of breast cancer.