Detecção do câncer de mama por meio de biópsia líquida: abordagens baseadas em espectroscopia infravermelha, metabolômica e inteligência artificial

Abstract

Introduction: Early and accurate detection of breast cancer (BC), the most frequently diagnosed cancer in women worldwide, remains a clinical challenge. Blood-based liquid biopsy (LB) represents a minimally invasive approach that allows not only cancer detection but also continuous monitoring of treatment response and disease progression. LB in cancer has been driven by powerful omics technologies, particularly attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy and liquid chromatography-mass spectrometry (LC-MS)-based metabolomics. Both enable the biochemical analysis of plasma and its circulating biomarkers, such as metabolites and extracellular vesicles (EVs), which may reflect tumor-derived molecular alterations. Objectives: To investigate the potential of ATR-FTIR, metabolomics, and artificial intelligence (AI)-based approaches applied to BC liquid biopsy by identifying molecular signatures in plasma, aiming at early detection and disease monitoring. Materials and methods: Plasma samples were collected from patients with non-metastatic BC (NMBC), metastatic BC (MBC), benign breast disease (BBD), and controls (CO). In the first article, plasma was analyzed by ATR-FTIR and global metabolomics. The potential of infrared signatures for clinical group prediction was assessed using machine learning (ML) models, while differential metabolites were identified using unpaired t-test. In the second article, plasma and plasma-derived EVs were analyzed by ATR-FTIR, and ML-based predictive models were also applied to the data. Prior to ATR-FTIR, EVs were isolated by ultracentrifugation and characterized by standard techniques. Results: In the first article, plasma ATR-FTIR spectroscopy presented high performance in distinguishing between CO and NMBC, and identified signatures associated with metastatic and benign conditions, although with lower performance in discriminating between benign and malignant. Metabolomic profiling corroborated the spectroscopic findings and revealed alterations in several metabolic pathways and potential biomarkers. In summary, it was identified a metabolic profile that differentiated BBD or NMBC from CO, another that differentiated BBD from NMBC, and one that was characteristic of MBC. In the second article, classification models based on EV spectra outperformed those based on plasma, particularly in distinguishing between NMBC and other groups (>80%), highlighting EVs as a more specific and informative source for LB. The inclusion of MBC reduced performance, reflecting greater heterogeneity associated with metastasis. As in the first article, plasma spectra showed high performance in distinguishing CO and BBD, a comparison of less clinical relevance but useful for characterizing benign breast conditions. Conclusion: This work demonstrated the potential of plasma and its circulating biomarkers, especially EVs, as promising tools for BL when analyzed using omics and AI techniques. These findings contribute to the development of more accurate, effective, and minimally invasive approaches for the early detection and monitoring of BC, reducing the need for invasive procedures.