A lab-on-a-bench setup equipped with machine learning for high-throughput formulation development

The conventional process of developing a formulation involves multiple steps, including individual formulation preparation, separate sample characterization, and empirical performance optimization, which is time, material, and effort consuming.

We have developed a lab-on-a-bench (LoB) setup equipped with Bayesian Optimization (BO) algorithm as an automated high-throughput method for the development of mixed oils, aiming to maximize surface wetting profile. LoB is an automated closed loop experimental platform that consists of a microfluidic chip as high throughput sample preparation part for mixing different proportions of jojoba oil and castor oil, followed by a quartz crystal microbalance (QCM) and a contact angle goniometer to quantify the rheological and interfacial characteristics of the prepared samples. BO, including both Gaussian process and Acquisition function, is deployed to predict the next most likely composition for the optimal performance based on the mean and covariance of the initial data, which feedback to the sample preparation.

The results showed that not only LoB and BO can run smoothly and reliably, but are able to predict the optimal ratio (0.61) between jojoba oil and castor oil based on five initial values, 0.59, 0.50, 0.44, 0.40, and 0.33 respectively, in producing the most desired wetting performance. We demonstrate that the LoB setup, equipped with the BO algorithm, offers a great potential to efficiently predict the optimal ratio of mixed oils in obtaining a maximum spreading ratio, which could be extended to the other complex formulated products.