Bayesian Optimization does not converge

I am trying to optimize a generative model to generate the RNA sequences with best scores which are the ON/OFF ratios of RNA toehold switches obtained from a neural network using Bayesian Optimization, but it does not converge. I implemented both BO and Multi Objective BO, but both of them does not converge. I could previously use BO for optimizing other features of the generative model, but for RNA toehold switches, it does not work. Is there any suggestions that how I can generate RNA toehold switches with best scores?