Machine Learning-Driven Dual-Recognition Magnetic Imprinted Polymers: Host–Guest/Aptamer Synergy Enabling Ultrasensitive Chloramphenicol Detection
Jiaming Zhang, Yanxia Ma, Jinbo Cao, Ai Li, Siying Liu, Xixiang Yang, Li Wang, Junhua Li, Xiaogang Hu
Abstract
To address the challenges in detecting chloramphenicol (CAP) in complex food matrices, this study developed a magnetic solid-phase microextraction coupled with a high-performance liquid chromatography (MSPME-HPLC) system that integrates machine learning and molecular recognition. The system employs magnetic SiO 2 @Fe 3 O 4 nanoparticles as the carrier and combines the dual recognition functions of carboxylated pillar[5]arene (CP[5]A) and aptamer (Apt) to create a nanocomposite separation material, Apt-MIP-CP[5]A@SiO 2 @Fe 3 O 4 (AC-MSF). Bayesian optimization and six machine learning models (e.g., XGBoost, SVM) were utilized to dynamically optimize polymerization and extraction conditions. SHAP interpretability analysis identified aptamer dosage and Mg 2+ concentration as critical parameters, with ML-recommended conditions reducing cross-linker usage by 22.2% and polymerization time by 57%. Kinetic simulations elucidated the synergistic recognition mechanism: CAP’s nitrobenzene group embeds in CP[5]A’s cavity via strong nonbonded interactions (total energy: −50 to −250 kJ/mol) and H-bond networks (1–4 bonds), while the aptamer binds CAP specifically at DT-18 via H-bonding (Δ G: −34.64 kcal/mol). Circular dichroism spectroscopy confirmed the independent yet synergistic operation of the dual recognition modes, with a synergy factor of 1.4. The system achieved a detection limit of 0.69 μg/L (linear range: 0.004–0.2 mg/L) and recoveries of 85.0%–96.2% in honey, milk, and egg samples.