Machine learning-based gene expression signature for classification of endocrine therapy sensitivity in ER+ breast cancer patients

Abstract

Abstract: Endocrine therapy (ET) combined with cyclin-dependent kinase 4/6 inhibitors (CDK4/6i) is the standard treatment for metastatic estrogen receptor-positive (ER+) breast cancer. However, not all patients require this combination therapy upfront, and identifying those who would respond well to ET alone could save healthcare costs, reserving the CDK4/6i combination for ET-resistant patients, and delay the eventual need for chemotherapy. In this study, we integrated two independent bulk RNA sequencing datasets, one inhouse generated and one publicly available, and used differential expression analysis (DEA) followed by LASSO selection to identify potential biomarkers associated with estrogen response. In this study, different machine learning techniques were employed to create predictive gene signatures, and comparisons were made based on performance. Through external validation with diverse datasets, we established a neural network-based 27-gene signature capable of classifying ET-sensitive patients with F-scores of up to 0.75. While validation presented challenges, our model offers promise for personalized clinical decision-making, provided more suitable validation data can be obtained.