Straightforward recursive partitioning model for discarding insoluble compounds in the drug discovery process

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Poor aqueous solubility is a significant challenge in drug discovery and development.
• Addressing solubility issues early in the research pipeline is essential for pharmaceutical success.

Purpose of the Study:

• To develop a predictive model for assessing compound aqueous solubility in early drug discovery.
• To determine if a compound is likely to be soluble enough for interpretable biological screening assay data.

Main Methods:

• Application of recursive partitioning (RP) method to 3563 molecules with in-house determined aqueous solubility values.
• Generation of five models using a limited set of descriptors related to structural features influencing solubility.
• Selection of a final model based on molecular weight (MW) and aromatic proportion (AP).

Main Results:

• The final model, utilizing only MW and AP, achieved 81% accuracy and 75% precision on a test set of 1200 compounds.
• The model provides intuitive insights into structural features affecting drug solubility.
• Successfully predicted sufficient solubility for interpretable biological screening data.

Conclusions:

• The developed model offers a valuable tool for compound selection and library design in early drug discovery.
• Predictive solubility assessment can mitigate development risks associated with poor solubility.
• This approach aids in optimizing the drug discovery process by identifying promising candidates early on.