Fig. 4

Probabilities of each of the 63 BsGDH-CatIB variants (plus soluble BsGDH as negative control) to be the best candidate. The probabilities are obtained using the sampling_probabilities function of the pyrff Python package, which essentially repeats Thompson sampling with several thousand samples from the probability distributions. The percentage in which a candidate was chosen by Thompson sampling is plotted as the probability to be the best variant in the library. Before any data is collected (Round 0, top), all variants are given the same prior probability for their reaction rate \({{\varvec{k}}}_{\mathrm{v}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{t}}\), which can be seen by the even distribution of probabilities. The number n above the bars indicates how often a variant was suggested by the algorithm for the upcoming round