Application of neural networks for response surface modeling in HPLC optimization S. Agatonovic-Kustrin a,* , M. Zecevic b , Lj. Zivanovic b , I.G. Tucker a a School of Pharmacy, University of Otago, PO BOX 913 Dunedin, New Zealand b Institute of Pharmaceutical Chemistry and Drug Analysis, Faculty of Pharmacy, Vojvode Stepe 450, 11000 Belgrade, Serbia Received 19 September 1997; received in revised form 11 December 1997; accepted 29 January 1998 Abstract The usefulness of arti®cial neural networks for response surface modeling in HPLC optimization is compared with multiple regression methods. The results show that neural networks offer promising possibilities in HPLC method development. The predicted capacity factors of analytes were better to those obtained with multiple regression method. # 1998 Elsevier Science B.V. Keywords: Optimization; Arti®cial neural networks; Multilinear stepwise regression analysis; Amiloride; Hydrochlorothiazide 1. Introduction The most important aspect in method development in liquid chromatography is achievement of suf®cient resolution in reasonable analysis time. This goal can be achieved by adjusting accessible chromatography factors, to give the desired response. A mathematical description of such goal is called an optimization. Usually, the methods focus on the optimization of the mobile phase composition, i.e. on the ratio of water and organic solvents (modi®ers). Besides, the optimi- zation of pH may lead to better selectivity. The degree of ionization of solutes, stationary phase and mobile phase additives may be affected by the pH. It is clear, however, that if the full power of eluent composition is to be realized, ef®cient strategies for multifactor chromatographic optimization must be developed. Retention mapping methods are useful optimization tool because the global optimum can be found [1]. The retention mapping is designed to completely describe or `map' the chromatographic behavior of solutes in the design space by its response surface, which shows the relationship between the response, i.e. the capacity factor of a solute, and several input variable, i.e. the components of the mobile phase. The capacity factor of every solute in the sample can then be predicted, rather than performing many separations and simply choosing the best one obtained [2]. Neural network methodology has found a rapidly increasing application in many areas of prediction both within and outside science. The main purpose of this study was to present the usefulness of arti®cial Analytica Chimica Acta 364 (1998) 265±273 *Corresponding author. Fax: 0064 3 479 7034; e-mail: nena.kustrin@stonebow.otago.ac.nz 0003-2670/98/$19.00 # 1998 Elsevier Science B.V. All rights reserved. PII S0003-2670(98)00121-4