Classification of signaling proteins based on molecular star graph descriptors using Machine Learning models

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Signaling proteins are vital for drug development, but their complex structures impede direct activity-structure correlation.
• Developing rapid, accurate, and cost-effective methods to evaluate novel molecular targets for diseases is a significant challenge.

Purpose of the Study:

• To develop a machine learning model for predicting signaling peptides using protein structure information.
• To overcome the limitations of directly associating signaling activity with complex protein structures.

Main Methods:

• Encoding peptide sequence information into topological indices using protein star graphs and the S2SNet tool.
• Employing Quantitative Structure-Activity Relationship (QSAR) classification models with Machine Learning techniques.
• Utilizing Support Vector Machines-Recursive Feature Elimination (SVM-RFE) with a Laplacian kernel (RFE-LAP) for model optimization.

Main Results:

• The best classification model, RFE-LAP, achieved an AUROC of 0.961, based on eleven descriptors.
• The model successfully predicted signaling peptides with high accuracy.
• Validation on 3114 proteins from the PDB database confirmed the model's predictive performance.
• Identified significant signaling pathways for three UniprotIDs with over 98.0% prediction accuracy.

Conclusions:

• The developed QSAR model effectively predicts signaling peptides, offering a valuable tool for drug discovery.
• This approach provides a faster, more accurate method for evaluating molecular targets compared to traditional methods.
• The findings facilitate the identification of novel signaling pathways and potential therapeutic targets.