Machine learning-based prediction of bioactivity in HIV-1 protease: insights from electron density analysis

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• HIV-1 protease is a critical target for antiviral therapy.
• Understanding molecular factors influencing inhibitor binding is essential for drug design.

Purpose of the Study:

• To develop a predictive model for HIV-1 protease inhibitor biological activity.
• To identify key molecular and structural factors governing enzyme inhibition.

Main Methods:

• Machine learning models were constructed using Atoms in Molecules theory and topological electron density analysis.
• Data included experimental X-ray crystallographic protein-ligand complexes and inhibition constants.
• Logistic regression was employed for classification and prediction.

Main Results:

• The logistic regression model achieved a test set accuracy of 0.76 and an AUC-ROC score of 0.77.
• Significant factors influencing activity included electron density contributions from hydrogen atoms and specific amino acid residues.
• The model demonstrated good differentiation between low and high activity compound classes.

Conclusions:

• The developed model effectively predicts HIV-1 protease inhibitor activity.
• Hydrogen bonding, glycine flexibility, and hydrophobic interactions are crucial for effective ligand binding.
• These insights aid in the rational design of novel HIV-1 protease inhibitors.