Leveraging-Induced Polarization for Drug Discovery: Efficient IC50 Prediction Using Minimal Features

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Atomic hybridization frequencies are key molecular descriptors.
• Molecular polarizability influences protein-ligand binding.
• Predictive models for drug efficacy are essential.

Purpose of the Study:

• To predict half-maximal inhibitory concentrations (IC50s) of drug-like molecules using atomic hybridization frequencies.
• To explore the role of induced polarization in protein-ligand binding.
• To assess the predictive power of atomic hybridization for molecular polarizability.

Main Methods:

• Utilized atomic hybridization frequencies (s, sp, sp2, sp3) as features for machine learning models.
• Employed Neural Network and Random Forest models for IC50 prediction.
• Correlated predicted molecular polarizabilities with atomic hybridization data.

Main Results:

• Achieved high correlation coefficients (R2) and low mean square error (MSE) using only 19 features.
• Demonstrated accurate prediction of molecular polarizabilities from atomic hybridization frequencies.
• Highlighted the significant role of induced polarization in protein-ligand binding.

Conclusions:

• Atomic hybridization frequencies are powerful predictors of IC50 values and molecular polarizability.
• Induced polarization's contribution to binding energy varies across different protein targets.
• This approach offers a novel method for drug discovery and development.