Machine learning-powered single-molecule cancer diagnosis using DNA origami tags.

Area of Science:

• Biophysics
• Nanotechnology
• Genomics

Background:

• Single-molecule detection (SMD) is crucial in biomedical research.
• Atomic force microscopy (AFM) offers nanoscale resolution for SMD but faces limitations in labeling and data processing speed.
• Machine learning (ML) presents an opportunity to overcome these AFM limitations.

Purpose of the Study:

• To develop an ML-powered strategy combining AFM and DNA nanotags for efficient SMD.
• To enable rapid and accurate cancer mutation detection using this novel approach.
• To demonstrate the practical applicability of the developed method in clinical diagnostics.

Main Methods:

• Utilized nickases for targeted DNA modification and insertion of distinct DNA nanotags.
• Employed AFM for high-resolution imaging of labeled DNA structures.
• Implemented a YOLOv5l algorithm for automated recognition and classification of AFM images.
• Validated the method on linear and circular DNA, and in patient samples with specific cancer mutations.

Main Results:

• The YOLOv5l algorithm achieved 98% accuracy in classifying 370 structures within 1.21 seconds.
• Successfully identified nickase-edited sites on DNA, confirming the proof of concept.
• Demonstrated high accuracy in detecting KRAS G12R and p53 R175H mutations in patient samples.
• Achieved diagnostic accuracy comparable to Sanger sequencing and quantitative PCR.

Conclusions:

• The developed ML-powered AFM strategy significantly enhances SMD capabilities.
• This method offers a rapid, accurate, and scalable solution for cancer mutation detection.
• The approach holds potential for advancing molecular diagnostics and biomedical research.