Triple-negative breast cancer (TNBC) lacks ER, PR and HER2 expression, represents ~10–20% of invasive breast cancers, and is clinically aggressive with limited targeted treatment options. Its pronounced molecular heterogeneity challenges single-marker diagnostics, motivating the development of robust biomarker panels for early detection and disease monitoring. Aptamers, and particularly guanine-rich DNA sequences forming G-quadruplexes (G4s), provide stable and versatile molecular recognition tools. Telomeric G4 structures are biologically relevant in genome maintenance and can enrich disease-related protein interactors, making them attractive baits for translational biomarker discovery. Here, we employed an overhang human telomere model capable of forming two consecutive G4s (tel46) immobilized on Controlled Pore Glass (CPG) to profile the nuclear G4 interactome in two TNBC cell lines, MDA-MB-231 and BT-549. Using affinity purification–mass spectrometry (AP-MS) with CPG-tel46 combined with quantitative proteomics and stringent background subtraction, we identified tel46-associated proteins consistently upregulated in both tumour models and prioritized 11 candidates supported by downstream bioinformatic validation. To move beyond single-marker evaluation, we implemented a machine-learning framework to assess candidate proteins as a coordinated molecular signature. Regularized models with embedded feature selection and cross-validation were used to identify stable, discriminative combinations while controlling overfitting. This integrative strategy supports G4-based capture as a practical approach to enrich clinically relevant interactors and prioritize diagnostic panels. We propose a five-protein signature (KIF4A, ACIN1, RBM12, FOXK1 and NCAPD2) as a candidate classifier for TNBC early diagnosis, providing a foundation for independent validation in clinical cohorts.