Batched Bayesian Optimization for Drug Design in Noisy Environments

Abstract

The early stages of the drug design process involve identifying compounds with suitable bioactivities via noisy assays. As databases of possible drugs are often very large, assays can only be performed on a subset of the candidates. Selecting which assays to perform is best done within an active learning process, such as batched Bayesian optimization, and aims to reduce the number of assays that must be performed. We compare how noise affects different batched Bayesian optimization techniques and introduce a retest policy to mitigate the effect of noise. Our experiments show that batched Bayesian optimization remains effective, even when large amounts of noise are present, and that the retest policy enables more active compounds to be identified in the same number of experiments.

Combinatorial mixtures of organic solutes for improved liquid/liquid extraction of ions

“…As the batch number increases, the regrets exhibit a declining trend, indicating that the predictions obtained using Bayesian optimization are getting closer to the optimal points. [41][42][43] We analyze the average variance for different mixtures in the Bayesian optimization method to have an estimation of the overall variation of the Bayesian optimization. The result shows that the average standard deviation is around 100 ppm, which is a relatively high variance (Fig.…”

“…However, with the help of Bayesian optimization we can handle the high variance and still find improved mixtures in a shorter time. [41][42][43] Fig. S7 (ESI †) gives us all the combinations and their measured values for each batch.…”

“…The details of the Bayesian optimization process used here are described elsewhere. [41][42][43] Briefly, a Python script was used to deal with the combinatorial optimization problem by searching the best mixture with the lowest outcome. To perform Bayesian optimization, we exploited the Gaussian process as a surrogate model to show the current uncertainties of all candidate combinations updated by n observed combinations in the first step.…”

Computational and data-driven modelling of solid polymer electrolytes

“…111 Bellamy et al used batch BO to explore a large database for use in drug design. 112 Specifically for the design of polymers, Li et al constructed ML surrogates for experiments and applied BO to propose short fiber polymer designs. 113 Gao et al also used an ML-based surrogate for the objective evaluation of BO for the design of polymeric membranes.…”

Large-Scale Pretraining Improves Sample Efficiency of Active Learning-Based Virtual Screening

“…Graff et al 29 developed a framework (MolPAL) based on batched Bayesian optimization, 30,31 which formulates a strong synergy with pool-based active learning, 28,32 to successfully recover 94.8% of the top-50,000 compounds from a 99.5 million-sized library by docking only 2.4% of it. Such a framework is shown to be effective in noisy environment 33 (a common problem with docking data), and a pruning algorithm is developed to improve efficiency by reducing the screening space. 34 The choice of the surrogate machine learning model in the active learning framework is one of the dominant factors that affects the hit recovery rate.…”