Bioactivity Prediction Based on Matched Molecular Pair and Matched Molecular Series Methods

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Lead optimization in small molecule drug discovery aims to enhance compound biological activity.
• Iterative synthesis and testing are laborious; high-quality in silico prediction is crucial to reduce trial-and-error.
• Accurate computational models can prioritize more active compound derivatives.

Purpose of the Study:

• To develop and evaluate in silico bioactivity prediction models.
• To compare models based on molecular similarity and SAR transferability.
• To identify the most effective modeling strategy for lead optimization.

Main Methods:

• Constructed two types of bioactivity prediction models using a large-scale structure-activity relationship (SAR) database.
• Developed similarity-based models (SA, SA_BR, SR, SR_BR) using matched molecular pair analysis.
• Developed SAR transferability-based models (Single MMS pair, Full MMS series, Multi single MMS pairs) using matched molecular series analysis.
• Defined model applicability domains using distance-based thresholds.

Main Results:

• The Multi single MMS pairs model demonstrated the best individual performance (R2 = 0.828).
• The baseline SA model showed lower prediction accuracy (R2 = 0.798).
• Consensus modeling further improved predictive accuracy, achieving R2 = 0.842, MAE = 0.397, and RMSE = 0.563.

Conclusions:

• An accurate bioactivity prediction model was developed using a consensus method.
• The consensus model outperformed all individual models, proving its superiority.
• This model serves as a valuable tool for accelerating lead optimization in drug discovery.