Enhancement of multianalyte mass spectrometry detection through response surface optimization by least squares and artificial neural network modelling.

In this work, the use of design of experiments and posterior data modelling by artificial neural network (ANN) and least squares (LS) is presented as a suitable analytical tool for the performance optimization of a tandem mass spectrometric detector coupled to ultra-high performance liquid chromatography for the analysis of seventeen veterinary drugs. Firstly, a central composite design was built considering as factors the cone, capillary, extractor and radio frequency voltages of the mass spectrometer in order to obtain a proper combination to improve the sensitivity of the method. Secondly, a one factor design considering the collision voltage was built to define the adequate voltage for each daughter ion. The response surface methodology (RSM) was then applied, and the prediction capability of ANN and LS were compared. As conclusion, the ANN modelling provided better results than LS, both in terms of the ANOVA and predicted areas results. The accuracy of the model prediction was between 85 and 125%, confirming that the estimates of the model were correct, and endorsing the optimization procedure as a suitable way to gather excellent results. The suitability of the new approach and its implications on the simultaneous analysis of seventeen veterinary drugs by ultra-high liquid chromatography coupled to tandem mass spectrometry detection are discussed.