Deep Learning Model for Fast Determination of Equilibrium Dissociation Constants Using Biolayer Interferometry...

Area of Science:

• Biochemistry
• Computational Biology
• Machine Learning

Background:

• Bio-Layer Interferometry (BLI) is crucial for detecting biomolecular interactions.
• Accurate prediction of equilibrium dissociation constant (K_d) is vital for understanding binding kinetics.
• Current methods may require extensive experimental conditions, limiting rapid analysis.

Purpose of the Study:

• To develop a deep learning model for rapid prediction of K_d values using BLI sensorgrams.
• To enhance model accuracy and generalization through SE and CBA module integration and fine-tuning.
• To create a comprehensive dataset of BLI curves for robust model training and validation.

Main Methods:

• A convolutional neural network (CNN) model was constructed for K_d prediction.
• YOLOv5 was utilized for automated extraction of 3812 BLI curves from literature.
• A dataset of 5640 sensorgrams was compiled, including lab-generated (wet) and GAN-generated (dry) data.

Main Results:

• The optimized deep learning model achieved 60% accuracy in predicting K_d values using wet data.
• The model successfully predicts K_d from single-concentration BLI sensorgrams.
• This method offers a viable alternative for rapid biomolecular interaction analysis.

Conclusions:

• Deep learning fused with BLI provides a powerful tool for predicting biomolecular binding affinity.
• The developed model enables efficient K_d prediction, especially in resource-limited scenarios.
• Future work with larger datasets will further improve prediction accuracy and model robustness.