Coupling analytical method validation with machine learning for mobile phase prediction in RP-HPLC

Mobile phase optimization in reversed-phase high-performance liquid chromatography (RP-HPLC) is conventionally performed through iterative trial-and-error approaches, resulting in substantial consumption of time, solvents, and active pharmaceutical ingredients (APIs). In this study, SolPred , a machine learning-based tool for predicting mobile phase compositions from molecular descriptors, was further developed using an expanded compound dataset to enhance predictive performance. The retrained model exhibited excellent accuracy and robustness ( r = 0.989, R² = 0.978, RMSE = 3.89, MAE = 2.44, CCC = 0.989). To ensure regulatory relevance, the predictive framework was integrated with the ICH Q2(R2) analytical method validation guidelines. The predicted mobile phase conditions were experimentally evaluated for five pharmaceutical compounds, with comprehensive validation performed for ciprofloxacin and dicloxacillin. All validation parameters met established acceptance criteria, including linearity (R² ≥ 0.997), accuracy (~ 100% recovery), precision (%RSD < 1%), and sensitivity (LOD < 0.5 µg mL⁻¹; LOQ < 1.5 µg mL⁻¹). Robustness was confirmed under deliberate variations in flow rate and detection wavelength. The proposed SolPred–Q2 framework demonstrates a reliable integration of machine learning prediction with regulatory-compliant validation, significantly reducing experimental workload and solvent consumption. This approach offers a practical and sustainable strategy for efficient RP-HPLC method development in pharmaceutical quality control.