Machine learning-based QSAR models to predict sodium ion channel (Na_v 1.5) blockers

Area of Science:

• Computational chemistry
• Cardiovascular pharmacology
• Drug discovery

Background:

• Drug-induced arrhythmias are a major safety concern.
• The voltage-gated sodium channel 1.5 (Na_v1.5) is a critical target for assessing proarrhythmic potential.
• Traditional experimental methods for evaluating proarrhythmic effects are costly and time-consuming.

Purpose of the Study:

• To develop machine learning models for predicting Na_v1.5 channel blockade by small molecules.
• To establish an efficient in silico screening method for early-stage drug discovery.

Main Methods:

• Development of machine learning classification and regression models.
• Utilizing datasets to train models for predicting channel blockade and pIC₅₀ values.
• Validation of model performance using balanced accuracies and q² scores.

Main Results:

• Classification models achieved high balanced accuracies (0.88-0.94) at different thresholds.
• The regression model accurately predicted pIC₅₀ with a q² of 0.84.
• These models demonstrate strong predictive power for Na_v1.5 channel interaction.

Conclusions:

• Machine learning models serve as effective filters for identifying potentially toxic compounds.
• In silico screening can significantly accelerate the drug discovery process.
• This approach aids in prioritizing safer drug candidates for further development.