Prediction of transient and permanent protein interactions using AI methods

Area of Science:

• Computational biology
• Biophysics
• Bioinformatics

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions and can be permanent or transient.
• Distinguishing between permanent and transient PPIs is vital for drug discovery and understanding biological mechanisms.

Purpose of the Study:

• To develop computational models for predicting and classifying protein-protein interactions as permanent or transient.
• To identify key features that discriminate between permanent and transient PPIs.

Main Methods:

• Calculated 43 physicochemical, geometrical, and structural features for 402 protein-protein complexes.
• Developed 5 supervised machine learning models using Scikit-learn and an Artificial Neural Network using TensorFlow and Keras.
• Evaluated model performance based on prediction accuracy.

Main Results:

• Machine learning models achieved prediction accuracies ranging from 76.54% to 82.71%.
• The k-Nearest Neighbors (k-NN) model demonstrated the highest accuracy.
• Interface area (e.g., Percent interface accessible area) and shape parameters (e.g., Planarity, Eccentricity) were identified as key discriminating factors.

Conclusions:

• The developed computational approach effectively predicts transient and permanent protein-protein interactions.
• This method offers a valuable, cost-effective alternative to experimental techniques for classifying PPIs.
• Understanding PPI classification aids in elucidating protein association mechanisms.