MicroRNA signatures and machine learning models for predicting cardiotoxicity in HER2 positive breast cancer patients

Abstract Background HER2-positive breast cancer (BC) patients receive as adjuvant treatment chemotherapy and targeted therapy, including anthracyclines and trastuzumab. At the same time, they face an increased risk of cardiotoxicity, which may lead to long-term cardiovascular complications. Identifying biomarkers that predict cardiotoxicity is crucial for implementation of cardioprotective therapies. MicroRNAs (miRNAs) have emerged as potential regulators of cardiotoxicity due to their role in gene expression and cardiovascular homeostasis. Aims This study aimed to evaluate the differential expression of circulating miRNAs in HER2-positive BC patients undergoing chemotherapy and targeted therapy and explore their predictive value for chemotherapy-induced cardiotoxicity using machine learning models. Methods A total of 47 BC patients were evaluated before and during chemotherapy. Blood samples were collected at baseline and three and six months after the first chemotherapy cycle. MiRNA expression profiling was performed using the miRCURY LNA miRNA PCR Panel, and differential expression was assessed using the 2−∆∆Ct method. Additionally, five machine learning models (decision tree, RF, SVM, GBM, kNN) were developed to classify cardiotoxicity based on miRNA expression levels. Results A total of 47 BC patients were included in the study, classified into two groups: those who developed cardiotoxicity and those who did not. Differential expression analysis identified 46 miRNAs with significant expression differences between groups (p0.05). Among them, 24 miRNAs were upregulated in the cardiotoxicity group, while others were significantly downregulated, with fold differences ranging from 14.66 to 26-fold (Figure 1). The machine learning models developed to assess the predictive value of miRNA expression profiles for cardiotoxicity classification had excellent performance (Figure 2). The decision tree model used hsa-miR-17-5p as the primary decision node, classifying samples with an expression level of ≥27 ΔCT as non-cardiotoxic with 100% certainty. In cases where hsa-miR-17-5p was 27 ΔCT, hsa-miR-185-5p served as a secondary classifier, further distinguishing cardiotoxic and non-cardiotoxic cases with high accuracy. According to SVM, hsa-miR-143-3p, hsa-miR-133b, hsa-miR-145-5p, hsa-miR-185-5p, and hsa-miR-199a-5p were associated with cardiotoxicity. According to the RF, hsa-miR-185-5p, hsa-miR-145-5p, hsa-miR-17-5p, hsa-miR-144-3p, and hsa-miR-133a-3p were associated with cardiotoxicity. Performance metrics revealed that kNN, SVM, and RF models outperformed the decision tree in overall predictive accuracy (Figure 1). Conclusion This study underscores the potential of circulating miRNAs as biomarkers for predicting cardiotoxicity in BC patients undergoing targeted therapy. Machine learning models could provide valuable insights into miRNA-based risk stratification, paving the way for personalized cardiotoxicity monitoring and intervention strategies.Figure 1 Figure 2