The data from this experiment was used to build an initial Kriging model of the concentration and pH dependence of PYK activity. In each of the remaining four rounds, the response-surface model was refined with 19 additional measurements. The parameters of these measurements were selected by our algorithm to lie in those regions where the lower bound of the model estimate is lowest, while keeping a user-specified minimum distance to other measurements. The minimum distance between points in the first refinement round was set to 0.25 with all parameters normalized to a interval. After each round, the minimum distance was multiplied in order to allow a finer sampling of the area around the estimated optimum. After each round of measurements, the response-surface was updated with the new results and its optimum was determined using R��s built in optimization method. The complete experiment was carried out autonomously without user interaction and ran for about two hours. Fig. 1 shows the measurements of the five iterations and the final response-surface model plotted together. The measurements of the later iterations are clustered around the optimum of the final response-surface model, providing much more information about this Omarigliptin region than 96 evenly distributed measurements could. The optimal parameters estimated after the first iteration already lay close to the centre of this region, showing that the algorithm is capable of finding useful reaction parameters even from minimal Cetylpyridinium chloride monohydrate number of experimental conditions tested if the range of conditions is selected correctly. The quality of the optimum estimate is somewhat hard to assess objectively because the true location of the optimum is unknown and the measurements are subject to a substantial amount of noise. To get some benchmark of the quality of the optimum in the response-surfaces for the individual iterations, we selected the five measurements closest to the estimated optimum and averaged their measured activities. The results in Table 1 show the activity of measurements close to the optimum increases from iteration to iteration, reflecting an improvement in the quality of the estimate.