Critical evaluation of AI-assisted HPLC method development: a case study of ranolazine analysis with systematic assessment of prediction accuracy and human optimization requirements

Abstract

This study introduces a parameter-level quantitative framework for the systematic evaluation of AI prediction accuracy in RP-HPLC method development ; a methodological contribution not previously established in the literature using ranolazine as a model analyte. Four AI platforms (Perplexity, Copilot, Gemini, and YesChat) were assessed using standardized chromatographic inputs. YesChat HPLC Method Developer was selected for in-depth evaluation for providing structured, actionable outputs with an explicit emphasis on green chemistry. The AI achieved an overall prediction accuracy of 72.9%, calculated as the mean of seven individually scored chromatographic parameters using a three-tier framework. Four instrumental parameters ; column temperature, flow rate, detection wavelength, and mobile phase ratio were predicted with 100% accuracy, while pH range scored 70%, as the AI suggested a broad range (3.0–4.5) rather than a specific value. Notable limitations emerged for chemical interaction parameters, particularly buffer necessity (0%) and cyclodextrin concentration (40%). Human-guided optimization subsequently removed the phosphate buffer and reduced sulfobutylether-β-cyclodextrin (SBE-β-CD) to 4 mmol/L, improving peak shape and retention (Rt = 6.17 min). The developed RP-HPLC method was validated according to ICH Q2(R1) guidelines and demonstrated robust analytical performance: linearity over 10.0–100.0 µg/mL (r² = 0.9999), precision with RSD < 2%, recovery of 98.66–101.20%, and LOD/LOQ of 1.25/3.79 µg/mL. Environmental performance was favorable, with an estimated WECA whiteness score of approximately 82/100, achieved by eliminating acetonitrile, reducing methanol consumption, and using cyclodextrin-assisted chromatographic optimization. These findings indicate that, within the scope of this evaluation, AI-assisted optimization can accelerate the development of sustainable, practical analytical methods for pharmaceutical quality control, provided that human expertise is applied to refine chemically complex parameters and ensure method robustness.