Fig. 1. Overview of machine learning enabled automated microfluidic design and control. (A) Complex microfluidic devices, such as the MATE-seq platform,⁹ is comprised of two components, a deterministic-lateral-displacement array and droplet generator, which can be parameterized to describe both the physical design and experimental conditions. By mapping these parameters to a target performance, datasets can be generated and used to train machine learning models that predict the performance of each component. Design automation tools can use these models to automatically design each component such that the desired performance is achieved while adhering to design constraints. Figure reproduced from Ng et al., 2019 (ref. 9) with permission from the Royal Society of Chemistry. (B) Microfluidic devices, such as a droplet sorter, can consist of a series of vital non-fluidic modules (schematic copyright 2017 National Academy of Sciences).¹⁰ These include: (1) flow controllers to drive behavior of the device; (2) sensors to measure and quantify occurring phenomena; (3) actuators to manipulate device behavior on the single-event level; and (4) feedback systems to respond to sensor information and intelligently control device behavior via the other modules. Machine learning provides a sophisticated feedback system to intelligently link modules together to perform complex tasks in real-time.