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

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Mitochondrial dysfunction is implicated in breast cancer development and progression.
• The nuclear genome's influence on mitochondria can drive cancer initiation.
• Limited research exists on mitochondrial genes as prognostic markers for early-stage breast cancer survival.

Purpose of the Study:

• To identify a reliable set of mitochondrial-related genes for predicting breast cancer patient survival.
• To develop a machine learning-based prognostic model for early-stage breast cancer.

Main Methods:

• Utilized machine learning (Lasso Cox regression, bootstrap) on bulk transcriptome data from the SCANB cohort.
• Analyzed 1136 known mitochondrial-related genes.
• Validated the model using the TCGA-BRCA external cohort.

Main Results:

• Identified a 14-gene mitochondrial signature predicting breast cancer survival (HR: 2.08, CI: 1.20-3.62).
• Developed a nomogram integrating the signature with clinical data for robust 1-, 3-, and 5-year survival prediction.
• Achieved strong predictive performance in both training (AUCs: 0.84, 0.79, 0.78) and validation cohorts (AUCs: 0.92, 0.83, 0.78).

Conclusions:

• A novel mitochondrial gene signature accurately predicts clinical prognosis in early-stage breast cancer.
• This signature holds potential as a valuable prognostic tool for patient management.