Modular microfluidic design automation using machine learning
Files
Main file
Date
2019-07-10
DOI
Authors
Lashkaripour, Ali
Rodriguez, Christopher
Mehdipour, Noushin
McIntyre, David
Densmore, Douglas
Version
Embargo Date
2019-07-08
OA Version
Citation
Lashkaripour, Ali, Christopher Rodriguez, Noushin Mehdipour, David McIntyre, and Douglas Densmore. "Modular microfluidic design automation using machine learning." 11th International Workshop on Bio-Design Automation (IWBDA-19), 2019.
Abstract
Microfluidics is the science of handling liquids inside sub-millimeter microchannels at nano-liter and pico-liter scales. This volume reduction enables increased resolution, sensitivity, and throughput, while, reducing the reagent cost significantly. These advantages make microfluidic devices to be ideal substitutes for bench-top and robotic liquid handling in numerous life science applications, specifically, synthetic biology where there is a need for low-cost, automated, and high-throughput platforms. Despite the need, implementing microfluidic platforms in the life science research work-flow is an exception rather than being the norm. This can be attributed to the high cost of fabricating microfluidic devices and a lack of microfluidic design automation tools that can design a microfluidic geometry based on the desired performance. As a result, designing a microfluidic device that delivers an expected performance is an iterative, time-consuming, and costly process. To address this, we previously described a low-cost and accessible micro-milling technique to fabricate microfluidic devices in less than an hour while costing less than $10. However, still designing a microfluidic device that performs as expected is an iterative and in-efficient process. Therefore, microfluidic design automation tools that are able to design a microfluidic geometry and provide the necessary flow conditions and fluid properties that would deliver a user-specified performance is with great importance. We propose a modular design automation tool, called DAFD, that is able to design a microfluidic device based on user-specified performance and constraints. DAFD uses machine learning to generate accurate predictive models, and then exploits these models to provide a design automation platform. DAFD can be implemented in many microfluidic applications such as droplet generation, high-throughput sorting, and micro-mixing.
Description
License
Attribution-NonCommercial 4.0 United States