FRET-SAM: SAM_Med2D-based automatic FRET two-hybrid analysis.

Area of Science:

• Biophysics
• Computational Biology
• Molecular Biology

Background:

• Fluorescence resonance energy transfer (FRET) two-hybrid assays quantify protein interactions in living cells.
• Manual analysis of FRET images is labor-intensive and computationally complex, limiting assay application.
• Deep learning offers a solution to automate FRET image analysis, improving efficiency and accuracy.

Purpose of the Study:

• To develop a deep learning method for automated analysis of FRET two-hybrid images.
• To enhance the efficiency, accuracy, and objectivity of FRET two-hybrid assays.
• To establish FRET-SAM as a tool for drug discovery by resolving drug-target interactions.

Main Methods:

• Developed FRET-SAM, an optimized analysis method based on the Segment Anything Model (SAM).
• Adapted SAM_Med2D for automated region of interest (ROI) selection and fluorescence signal extraction in FRET images.
• Trained and validated FRET-SAM using a comprehensive FRET image dataset with multiple plasmids and FRET pairs.

Main Results:

• FRET-SAM demonstrated improved segmentation accuracy (MPA, MIoU, Dice coefficient) compared to SAM_Med2D.
• FRET-SAM-derived results showed high consistency with literature values (relative errors < 5%).
• FRET-SAM showed potential in drug screening case studies, resolving drug-target interactions.

Conclusions:

• FRET-SAM significantly enhances FRET two-hybrid assay efficiency and accuracy through automated image analysis.
• The elimination of subjective bias improves the reliability of protein interaction quantification.
• FRET-SAM is a promising tool for drug discovery due to its ability to resolve drug-target interactions.