Toward Dose Prediction at Point of Design.

Area of Science:

• Drug discovery and development
• Computational chemistry
• Pharmacokinetics

Background:

• Human dose prediction (HDP) is crucial for optimizing compounds in preclinical drug discovery.
• Accurate HDP can streamline the identification of viable drug candidates.
• Current methods may require significant experimental data early in the process.

Purpose of the Study:

• To develop an exclusively in silico HDP strategy for triaging compound designs.
• To create a model estimating human dose based on chemical structure for a target exposure.
• To assess and incorporate model uncertainty for reliable dose estimations.

Main Methods:

• Machine learning models were built to estimate rat pharmacokinetics.
• Rat pharmacokinetics were allometrically scaled to predict human pharmacokinetics.
• A 10 nM free concentration target was used for early HDP in the absence of potency data.
• Uncertainty analysis was performed and propagated to the final dose estimate.

Main Results:

• The in silico HDP strategy can reduce the synthesis of compounds with unfavorable properties compared to unstructured approaches.
• The model provides a preliminary dose estimate based on chemical structure.
• Uncertainty quantification offers guidance on the reliability of the HDP estimates.

Conclusions:

• The developed in silico HDP strategy offers a valuable tool for early-stage compound optimization.
• This computational approach aids in efficient preclinical drug discovery by prioritizing compounds.
• While beneficial, the strategy necessitates continued experimental validation and testing.