Graph neural network modeling of receptor interaction kinetics from single-molecule imaging data

Area of Science:

• Biophysics
• Computational Biology
• Cell Biology

Background:

• Single-molecule imaging (SMI) captures live-cell receptor interactions crucial for cell signaling.
• Current SMI methods often provide incomplete interaction data due to substoichiometric labeling.

Purpose of the Study:

• To develop a computational method, Deep-FISIK, for predicting full receptor system kinetics from SMI data.
• To overcome limitations of partial labeling in SMI for studying receptor interactions.

Main Methods:

• Developed Deep-FISIK, a graph neural network model with multi-head attention for message-passing.
• Utilized single-molecule detections from SMI experiments as input, without requiring explicit molecular tracking.
• Designed for compatibility with higher receptor labeling fractions in SMI experiments.

Main Results:

• Deep-FISIK accurately predicts the kinetics of homotypic receptor interactions for the entire system.
• The method enhances prediction accuracy of interaction kinetics parameters compared to traditional SMI analysis.
• Demonstrated robustness of Deep-FISIK across various deviations from training data.

Conclusions:

• Deep-FISIK enables more comprehensive analysis of receptor interactions using SMI data.
• The computational approach improves the study of cell signaling dynamics by leveraging partial labeling.
• Deep-FISIK is applicable to diverse receptor systems and SMI experimental conditions.