Perspective
Toward microfluidic continuous-flow and intelligent downstream processing of biopharmaceuticals
The concept of miniaturized biopharmaceutical downstream processing with AI-controlled continuous flow platforms is described to overcome limitations of conventional processes, potentially accelerating the development of novel biotherapeutics.
High-throughput microfluidic systems accelerated by artificial intelligence for biomedical applications
This review outlines the current advances of high-throughput microfluidic systems accelerated by AI. Furthermore, the challenges and opportunities in this field are critically discussed as well.
High-speed cell partitioning through reactive machine learning-guided inkjet printing
We present a high-throughput single cell dispensing instrument using an inkjet printhead under real-time optical monitoring, guided by machine learning feedback. The instrument rapidly partitions cells for assays such as in whole genome sequencing.
Paper
Advancing microfluidic design with machine learning: a Bayesian optimization approach
The proposed Bayesian optimization-based approach enhances micromixer performance by optimizing geometric parameters, significantly reducing required number of simulations, and accelerating the design process compared to conventional methods.
Wearable intelligent sweat platform for SERS-AI diagnosis of gout
A wearable intelligent SERS platform enables the sensitive detection of UA (0.1 μM) in sweat for AI diagnosis of gout.
Angio-Net: deep learning-based label-free detection and morphometric analysis of in vitro angiogenesis
This study introduces Angio-Net, which integrates a high-throughput 3D cell culture device, large-scale image data generation, and deep learning-based virtual staining. The system demonstrated fast and accurate quantitative analysis of complex angiogenesis.
Paper
Label-free cell classification in holographic flow cytometry through an unbiased learning strategy
Unbiased learning pipeline for label-free single-cell classification.
About this collection
Handpicked by our Associate Editor, Hang Lu (Georgia Tech), we are pleased to highlight select works on artificial intelligence published in recent years. Read what she had to say below:
Artificial intelligence and microfluidics have entered a phase of productive convergence. What began as isolated demonstrations (e.g. neural networks classifying droplet images, genetic algorithms optimizing channel geometries) has matured into a bidirectional relationship. AI methods now enhance how we design, operate, and interpret microfluidic systems, while microfluidic platforms generate the high-throughput, standardized data that machine learning demands. This Editor's Choice collection highlights recent work at this interface, spanning device design, image-based analysis, and integrated diagnostic and processing systems.
A natural starting point is the comprehensive review by Zhou et al., which surveys the convergence of high-throughput microfluidic systems and artificial intelligence across biomedical applications (10.1039/D3LC01012K). The review addresses a core tension in the field: while automation and throughput have advanced rapidly, the volume of data generated by modern microfluidic platforms increasingly outpaces the capacity for manual analysis. The authors provide an accessible introduction to relevant AI methods while examining applications in biomedical detection, drug screening, and automated system control and design. The review also offers a discussion of current challenges and future opportunities, making it a valuable orientation for researchers entering this space or seeking to identify gaps where new contributions would have impact.
On the design front, Kundacina et al. demonstrate a Bayesian optimization framework for microfluidic mixer development, achieving optimal geometries at least an order of magnitude faster than conventional optimization methods while eliminating the need for separate surrogate models (10.1039/D4LC00872C). This approach of systematic exploration of design space with minimal simulations generalizes readily to other device types such as droplet generators and particle separators. Cheng et al. push automation further with the Isolatrix, an inkjet-based instrument that combines continuous optical feedback with machine learning classification for high-speed single-cell partitioning (10.1039/D5LC00514K). The system achieves a very high classification accuracy at orders of magnitude the speed of manual operation, enabling large-scale genomic profiling workflows including single-cell whole genome sequencing.
Image-based analysis remains a natural fit for deep learning. Ciaparrone et al. report label-free cell classification in holographic flow cytometry, addressing a key challenge: learning-based models often suffer from biases when trained on data from specific imaging settings (10.1039/D3LC00385J). Their approach combines Mask R-CNN for cell detection with a convolutional auto-encoder operating on unlabelled data, enabling generalization across experimental conditions. Kim et al. introduce Angio-Net, which replaces conventional immunocytochemistry with deep learning-based virtual staining, converting brightfield images into label-free fluorescence images (10.1039/D3LC00935A). Integrated with a standardized microfluidic device, the system enables high-throughput quantitative angiogenesis analysis without cell fixation, which could be valuable for drug efficacy screening and tumor microenvironment studies.
We also highlight applications where AI is embedded within microfluidic devices. Chen et al. present a wearable platform coupling SERS with AI for gout diagnosis (10.1039/D3LC01094E), Hormozinezhad et al. combine dimensional analysis with machine learning for non-Newtonian droplet generation (10.1039/D4LC00946K), and Dong et al. outline a framework for intelligent downstream processing of biopharmaceuticals (10.1039/D3LC01097J).
Looking ahead, we see substantial opportunities for deeper integration, particularly in closed-loop process control, where AI-driven decision-making operates continuously rather than at discrete endpoints, and in physics-informed learning approaches that embed domain knowledge into model architectures. We encourage submissions that push beyond proof-of-concept demonstrations toward robust, generalizable systems validated across diverse experimental conditions.
We hope this collection illustrates the expanding scope of AI–microfluidics integration and inspires further contributions to Lab on a Chip.