Machine Learning-Driven Capillary Microfluidic Design Automation for Programmable Gradient Generation and Antimicrobial Testing.

Rapid, portable, and automated determination of minimum inhibitory concentration (MIC) is critical for antimicrobial susceptibility testing in decentralized settings. We present TCG-CMDA (Tree-Shaped Concentration Gradient generator enabled by Capillary Microfluidic Design Automation), a machine learning-guided platform for the automated design of capillary microfluidic chips that passively generate programmable concentration gradients. A key innovation of TCG-CMDA is its ability to achieve fully synchronized, capillary-driven flow of two preloaded agents, enabling precise and reproducible gradient formation without external pumps or valves. By integrating computational fluid dynamics (CFD) simulations, a neural network surrogate model, and quasi-Newton optimization, the platform customizes tree-shaped geometries to match target mixing profiles. We validate TCG-CMDA through dye flow visualization and MIC testing of Escherichia coli against gentamicin, showing high concordance with conventional antibiotic susceptibility testing. The resulting chip supports a streamlined workflow consisting of passive loading, autonomous gradient generation, and incubation (∼2 h), followed by direct endpoint assessment, making it well suited for decentralized point-of-care applications. TCG-CMDA establishes a closed-loop, simulation-guided design paradigm for capillary microfluidic systems. Although demonstrated here for MIC-oriented antimicrobial testing, the framework is generalizable and can be extended to other gradient-dependent microfluidic applications through geometry reconfiguration and fluid-specific optimization.