Application of artificial neural network for simultaneous analysis of active pharmaceutical ingredients in multicomponent drug formulation

Abstract

Objectives: To optimize Artificial Neural Network (ANN) models for the simultaneous determination of active ingredients in pharmaceutical formulations and to apply these models to the simultaneous analysis of active ingredients in multicomponent drug products. Methods: Standard solutions of Metronidazole (MTZ), Spiramycin (SPI), Ezetimibe (EZE), Atorvastatin calcium (ATO), Bisoprolol fumarate (BIS), and Hydrochlorothiazide (HCT) were prepared in methanol to obtain appropriate stock solutions. Mixed standard samples were constructed using full factorial and central composite designs, and UV-Vis spectra were scanned over wavelength ranges appropriate to each preparation. A multilayer feedforward ANN model, using the Adam optimizer and ReLU activation function, was optimized for the number of input wavelengths, number of hidden layers, number of neurons, and learning rate; performance was evaluated using RMSEC, RMSECV, and R². The method was validated according to AOAC guidelines, with results compared to HPLC or microbiological assay using t-tests. Results: The optimized models achieved R² values of 0.9945–0.9999 and RMSECV values of 0.0744–0.7194. Recovery rates ranged from 90–107%, and RSD% for repeatability and intermediate precision met acceptance criteria. Quantification results by ANN showed no statistically significant difference from reference methods. Conclusion: ANN models demonstrated accuracy and reliability comparable to the reference methods. The proposed approach has potential for broad application in pharmaceutical quality control laboratories, reducing both analysis cost and time while maintaining analytical reliability.