Support Vector Models-Based Quantitative Structure–Retention Relationship (QSRR) in the Development and Validation of RP-HPLC Method for Multi-component Analysis of Anti-diabetic Drugs

Abstract

This work emphasized the use of the quantitative structure–retention relationship (QSRR) approach in the prediction retention time of anti-diabetic drugs on C₁₈ column in the HPLC method development process. This in silico QSRR study utilized a data set from literature and in-house studies for the development of better predictive model. A total of 11 QSRR models were developed and narrowed to 5 candidate models using a mobile phase composition range. The candidate models 1, 2, 3, 4, and 5 showed R² scores of 0.8844, 0.8968, 0.8996, 0.9769, and 0.9916, respectively. The model validation data revealed that support vector model (SVM)-based models 4 and 5 showed better predictive ability (> 99%) than the random forest model. The R² value for capacity factor prediction for models 4 and 5 was 0.862 and 0.881, respectively. Accordingly, the experimental retention time of pioglitazone, glimepiride, gliclazide, glyburide, and metformin was experimentally verified. Accordingly, we demonstrated good correlation (R² > 0.9) between experimental and predictive retention time on C₁₈ column. Based on prediction, a new HPLC method was optimized for the simultaneous analysis of pioglitazone (3.6 ± 0.2 min) and glimepiride (6.1 ± 0.2 min) on C₁₈ column using a mobile phase consisting of methanol and 0.1% ortho phosphoric acid (pH 2.7) with detection at 227 nm. The respective % retention prediction error was 0.2% and 6.3% for pioglitazone and glimepiride. The method demonstrated the linearity with regression coefficients of 0.9985 and 0.9998, respectively, for pioglitazone (15–75 µg/mL) and glimepiride (2–10 µg/mL). The % RSD (0.77–1.43%) and % accuracy (98.01–102.39%) of the method were acceptable. The method has proven specificity in the presence of degradation products and demonstrated robustness (< 2%RSD) to critical method parameters.