Machine Learning Models to Predict Protein-Protein Interaction Inhibitors.

Area of Science:

• Drug discovery and development
• Computational chemistry
• Bioinformatics

Background:

• Protein-protein interaction (PPI) inhibitors are crucial in modern drug discovery.
• Machine learning (ML) offers a promising approach for identifying and classifying potential PPI inhibitors.

Purpose of the Study:

• To evaluate the performance of various ML algorithms and molecular fingerprints for classifying PPI inhibitors.
• To develop and provide freely accessible computational tools for medicinal chemists.

Main Methods:

• Comparison of classification algorithms including Random Forest (RF), Logistic Regression (LR), and Support Vector Machines (SVM).
• Evaluation of different molecular fingerprints: Extended Connectivity Fingerprint radius 2 (ECFP4), ECFP6, and MACCS keys (166-bits).
• Development of ensemble models based on top-performing individual models.

Main Results:

• Random Forest (RF) models utilizing ECFP4 achieved the highest classification accuracy compared to other fingerprints and algorithms.
• While LR models generally showed lower performance, ECFP4 proved to be the most suitable representation.
• SVM models also performed well with the ECFP4 representation.

Conclusions:

• The study highlights the effectiveness of ML, particularly RF with ECFP4, for identifying PPI inhibitors.
• Freely available pipeline code is provided to assist researchers, especially non-computational experts, in drug discovery efforts.
• The findings contribute to advancing the field of computational drug discovery for PPI targets.