Rapid DNA origami nanostructure detection and classification using the YOLOv5 deep convolutional neural network.

Area of Science:

• Nanotechnology
• Biophysics
• Computer Science

Background:

• Accurate identification of DNA nanostructures is crucial for DNA origami prototyping and quality control.
• Current methods can be time-consuming and may lack scalability for large datasets.

Purpose of the Study:

• To adapt the YOLO object detection platform for high-fidelity identification of DNA nanostructures within atomic force microscope (AFM) images.
• To create a widely accessible, open-source tool for automated DNA structure analysis.

Main Methods:

• DNA origami structures (triangles and breadboards) were designed and fabricated.
• Atomic force microscopy (AFM) was used to image the DNA structures.
• Data annotation, augmentation, and YOLO model training were performed for single-class object detection.
• A sequential application of trained models was used for identification in complex image datasets.

Main Results:

• Two distinct YOLO models were trained, one for each DNA origami architecture.
• The system successfully identified 3470 out of 3617 DNA structures across various image conditions, including impurities.
• Analysis was completed in under 20 seconds, achieving an F1 score of 0.96.

Conclusions:

• The YOLO object detection platform offers a robust and efficient solution for intra-image identification of DNA nanostructures.
• This approach facilitates rapid prototyping and quality control in DNA origami research.
• The developed open-source system is adaptable for diverse structural geometries and imaging modalities beyond AFM.