Integrative Machine Learning Model for Overall Survival Prediction in Breast Cancer Using Clinical and Transcriptomic...

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Luminal A breast cancer, typically favorable, can be influenced by age and menopausal status.
• Conventional prognostic models may not fully capture survival nuances in diverse age groups.

Purpose of the Study:

• To investigate age-related transcriptomic differences in Luminal A breast cancer.
• To develop an advanced prognostic model integrating clinical and genomic data for improved survival prediction.

Main Methods:

• Analysis of transcriptomic and clinical data from the METABRIC cohort, stratifying patients by age and menopausal status.
• Identification of differentially expressed genes (DEGs) and feature selection using Boruta.
• Training and validation of machine learning models (Random Forest, Logistic Regression, Multilayer Perceptron, XGBoost) with cross-validation and SMOTE.

Main Results:

• Distinct transcriptomic clustering observed across different age groups.
• Identification of 41 age- and survival-associated genes, with key predictors including clinical variables and molecular markers (e.g., ATM, HERC2).
• XGBoost model achieved high performance (accuracy 98%, AUC 0.86), outperforming other algorithms.

Conclusions:

• Age-related transcriptomic alterations significantly affect Luminal A breast cancer prognosis.
• An integrated machine learning approach combining clinical and molecular data enhances prognostic accuracy.
• This ML-based strategy shows potential for clinical application in personalized breast cancer management.