A Machine Learning Model to Predict the Triple Negative Breast Cancer Immune Subtype

Area of Science:

• Oncology
• Immunotherapy
• Genomics

Background:

• Immune checkpoint blockade (ICB) therapy improves progression-free survival (PFS) in triple-negative breast cancer (TNBC).
• However, only about 10% of TNBC patients achieve complete response (CR) due to low response rates and adverse reactions.
• There is a need to identify biomarkers for predicting ICB response and patient prognosis in TNBC.

Purpose of the Study:

• To identify novel immune subtypes in triple-negative breast cancer (TNBC) based on gene expression profiles.
• To evaluate the prognostic value of these subtypes and their correlation with immune checkpoint blockade (ICB) response.
• To develop a predictive tool for stratifying TNBC patients for ICB therapy.

Main Methods:

• Unsupervised clustering analysis was performed on open datasets (TCGA, GEO) of 931 TNBC samples to identify immune subtypes.
• Prognosis, pathway enrichment, and ICB indicators were compared between subtypes.
• Machine learning models, including random forest, were used to build a predictive web server (TNBCIS) for immune subtype classification.

Main Results:

• Two novel immune subtypes, S1 and S2, were identified in TNBC.
• Subtype S1 exhibited higher immune scores, increased immune cell infiltration, and a significantly better prognosis (OS, RFS) compared to S2.
• A random forest model based on 11 hub genes achieved an AUC of 0.76 for predicting subtypes, and the TNBCIS web server was developed.

Conclusions:

• TNBC can be stratified into distinct immune subtypes with differential prognoses and immunotherapy responses.
• The identified subtypes and the predictive TNBCIS web server can aid in selecting TNBC patients likely to benefit from ICB.
• This approach facilitates personalized treatment strategies for improved outcomes in TNBC patients undergoing immunotherapy.