Integrating real-time deep learning for automation of optical tweezers experiments

The perhaps most widely used tool for measuring forces and manipulating particles at the micro and nano-scale are optical tweezers which have given them widespread adoption in physics, chemistry and biology. Despite advancements in computer interaction driven by large-scale generative AI models, experimental sciences—and optical tweezers in particular—remain predominantly manual and knowledge-intensive, owing to the specificity of methods and instruments. Here, we demonstrate how integrating the components of optical tweezers—laser, motor, microfluidics, and camera—into a single software simplifies otherwise challenging experiments by enabling automation through the integration of real-time analysis with deep learning. We highlight this through a DNA pulling experiment, showcasing automated single molecule force spectroscopy and intelligent bond detection, and an investigation into core-shell particle behavior under varying pH and salinity, where deep learning compensates for experimental drift. We conclude that automating experimental procedures increases reliability and throughput, while also opening up the possibility for new types of experiments.