Abstract
Background: Breast cancer is one of the leading causes of cancer-related mortality among women worldwide. Despite advancements in treatment, therapeutic resistance remains a major challenge, necessitating novel approaches for more effective interventions. One of the hallmarks of cancer, particularly in breast tumors, is metabolic reprogramming, where altered metabolic pathways create distinct profiles compared to normal cells. Identifying these metabolic alterations can provide critical insights for developing targeted therapies aimed at disrupting tumor metabolism and improving patient outcomes.
Objectives: This study applies six machine learning algorithms to predict metabolic profiles in breast cancer patients compared to healthy individuals, providing a promising approach for identifying metabolic targets in precision therapy.
Method: Plasma samples from 102 breast cancer patients and 99 healthy individuals were analyzed using targeted liquid chromatography-tandem mass spectrometry (LC-MS/MS) to assess metabolic profiles. Six machine learning algorithms were applied to evaluate classification performance, and feature importance was determined using the Mean Squared Error (MSE) value.
Result: Our findings revealed a significant decrease in alanine, histidine, tryptophan, tyrosine, methionine, and proline levels in breast cancer patients. Among the machine learning models, Random Forest (RF) achieved the highest classification performance (accuracy: 0.90, specificity: 0.85, sensitivity: 0.95), followed by K-Nearest Neighbors (KNN) with similar sensitivity but lower specificity. Logistic Regression (LR) balanced specificity (0.90) and sensitivity (0.86) with an accuracy of 0.88. Naïve Bayes (NB) and Support Vector Machine (SVM) showed moderate accuracy (0.83), while Decision Tree (DT) had the lowest sensitivity (0.76) but the highest PPV (0.89). Feature importance analysis identified glutamic acid, ketocholesterol, cystine, ornithine, succinate, acetylcarnitine, asparagine, tryptophan, and palmitic acid as key metabolic markers.
Conclusion: This study draws attention to key predictive metabolic bottlenecks identified through machine learning models, which could aid in targeted therapy and personalized treatment based on patients' metabolic profiles.
Keywords
Breast cancer, Metabolic alterations, Classification algorithms, Metabolic vulnerability, Feature importance analysis