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

Area of Science:

• Biotechnology
• Microfluidics
• Artificial Intelligence

Background:

• Single-cell analysis is vital for understanding cellular heterogeneity.
• Existing single-cell sample manipulation (SCSM) systems are costly, low-throughput, and labor-intensive.
• Limited research applies large language models (LLMs) to lab-on-a-chip automation.

Purpose of the Study:

• To develop a fully automated SCSM platform using an active-matrix digital microfluidic (AM-DMF) system.
• To integrate LLMs and object detection for intelligent SCSM.
• To create a novel, programmable lab-on-a-chip system for automated biological procedures.

Main Methods:

• Development of an active-matrix digital microfluidic (AM-DMF) platform.
• Integration of large language models (LLMs) and object detection algorithms.
• Implementation of a three-class detection method for cell and oil bubble identification, including edge cases.

Main Results:

• Achieved a single-cell sample generation rate of up to 25% and identification precision of 98%.
• Improved cell recognition accuracy by 1.0% (AP75 metric) by distinguishing cells at droplet edges.
• Demonstrated efficient handling of obscured cells at droplet edges, improving overall SCSM performance.

Conclusions:

• The developed AM-DMF platform offers a breakthrough in automated SCSM, surpassing existing systems in efficiency and performance.
• The integration of LLMs provides a ubiquitous tool for automatic SCSM workflow generation, advancing single-cell analysis.
• This work paves the way for more accessible and efficient single-cell studies in life sciences.