Computer vision techniques for high-speed atomic force microscopy of DNA molecules.

Area of Science:

• Nanotechnology
• Genomics
• Biophysics

Background:

• High-speed atomic force microscopy (HSAFM) generates vast nanoscale image datasets.
• Analyzing these images for single DNA molecule detection is crucial for genetic diagnostics but is labor-intensive.
• Current manual analysis presents a bottleneck in processing HSAFM data.

Purpose of the Study:

• To investigate the application of deep learning for streamlining HSAFM image analysis in genetic testing.
• To develop and compare machine learning models for automated DNA molecule detection and classification.
• To improve the efficiency and accuracy of diagnosing genetic disorders using HSAFM data.

Main Methods:

• Implemented a fully convolutional network (FCN) for image quality assessment of trinucleotide repeat expansion disease samples.
• Utilized the YOLOv8 object detection architecture to identify marked DNA molecules from Fragile X syndrome patients.
• Compared FCN performance against traditional methods like Laplacian of Gaussian and fast Fourier transform.

Main Results:

• The FCN achieved 96% accuracy and an AUC of 0.990 in reproducing human categorizations of image quality.
• The YOLOv8 model demonstrated an average precision of 0.966 in detecting marked DNA molecules.
• The object detection model successfully identified target DNA molecules within a large dataset, significantly reducing analysis time.

Conclusions:

• Deep learning methods, specifically FCN and YOLOv8, effectively automate and enhance HSAFM data analysis for genetic diagnostics.
• Machine learning integration promises to accelerate sample analysis and improve diagnostic precision in genomics.
• This approach offers a powerful tool for rapid identification of genetic markers in disease research and clinical applications.