GRAM: A True Null Model for Relative Binding Affinity Predictions.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Accurate relative binding affinity prediction is crucial for computer-aided drug design (CADD).
• Existing in silico methods lack robust performance assessment due to the absence of a true null model for objective comparison.
• Standard metrics like Pearson correlation coefficient, MUE, and RMSE do not provide a reliable baseline for random prediction performance.

Purpose of the Study:

• To introduce a novel performance measure, the prediction interval (PI), for assessing relative binding affinity prediction methods.
• To establish a true and nontrivial null model for objective evaluation of in silico prediction tools.
• To enable standardized comparison of different prediction algorithms and measure progress in the field.

Main Methods:

• Introduction of prediction intervals (PI) estimated from the error distribution of predictions.
• Development of the Gaussian Random Affinity Model (GRAM) as a null model.
• GRAM is based on the normal distribution of affinity changes observed in lead optimization (N(0, σ)).

Main Results:

• Prediction intervals (PI) offer an uncertainty range for predicted activities, vital for prospective applications.
• The Gaussian Random Affinity Model (GRAM) provides an analytically defined PI dependent only on activity variation.
• GRAM establishes a well-defined null model for objective performance assessment.

Conclusions:

• The proposed prediction interval (PI) and Gaussian Random Affinity Model (GRAM) address the need for robust performance assessment in relative binding affinity prediction.
• This approach facilitates standardized comparison of computational methods, crucial for advancing drug design algorithms.
• The developed framework supports objective evaluation and progress measurement in the field of in silico drug design.