Intelligent single-cell manipulation: LLMs- and object detection-enhanced active-matrix digital microfluidics

Abstract Single-cell analysis is crucial for deciphering cellular heterogeneity and understanding complex biological systems. However, most existing single-cell sample manipulation (SCSM) systems suffer from various drawbacks such as high cost, low throughput, and heavy reliance on human interventio...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhiqiang Jia, Chen Jiang, Jiahao Li, Yacine Belgaid, Mingfeng Ge, Li Li, Siyi Hu, Xing Huang, Tsung-Yi Ho, Wenfei Dong, Zhiwen Yu, Hanbin Ma
Format: Article
Language:English
Published: Nature Publishing Group 2025-07-01
Series:Microsystems & Nanoengineering
Online Access:https://doi.org/10.1038/s41378-025-00962-y
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Single-cell analysis is crucial for deciphering cellular heterogeneity and understanding complex biological systems. However, most existing single-cell sample manipulation (SCSM) systems suffer from various drawbacks such as high cost, low throughput, and heavy reliance on human interventions. Currently, large language models (LLMs) have been used in robotic platforms, but a limited number of studies have reported the application of LLMs in the field of lab-on-a-chip automation. Consequently, we have developed an active-matrix digital microfluidic (AM-DMF) platform that realizes fully automated biological procedures for intelligent SCSM. By combining this with a fully programmable lab-on-a-chip system, we present a breakthrough for SCSM by combining LLMs and object detection technologies. With the proposed platform, the single-cell sample generation rate and identification precision reach up to 25% and 98%, respectively, which are much higher than the existing platforms in terms of SCSM efficiency and performance. Furthermore, a three-class detection method considering droplet edges is implemented to realize the automatic identification of cells and oil bubbles. This method achieves a 1.0% improvement in cell recognition accuracy according to the $${\rm{AP}}_{75}^{\rm{test}}$$ AP 75 test metric, while efficiently distinguishing obscured cells at droplet edges, where approximately 20% of all droplets contain cells at their edges. More importantly, as the first attempt, a ubiquitous tool for automatic SCSM workflow generation is developed based on the LLMs, thus advancing the development and progression of the field of single-cell analysis in the life sciences.
ISSN:2055-7434