Aqueous solubility prediction based on weighted atom type counts and solvent accessible surface areas.

Area of Science:

• Computational chemistry
• Drug discovery
• Quantitative structure-property relationship (QSPR) studies

Background:

• Accurate prediction of aqueous solubility is crucial for drug development.
• Existing models may lack generalizability across diverse chemical structures.
• Developing robust quantitative structure-property relationship (QSPR) models is essential for efficient drug discovery pipelines.

Purpose of the Study:

• To develop and validate reliable aqueous solubility models for a large and diverse dataset of compounds.
• To identify the most predictive model for assessing druglikeness and prioritizing compound libraries.
• To provide tools for computational drug discovery and high-throughput screening (HTS).

Main Methods:

• Development of four aqueous solubility models: ASM-ATC, ASM-ATC-LOGP, ASM-SAS, and ASM-SAS-LOGP.
• Utilizing atom type counts, ClogP, and solvent accessible surface areas as molecular descriptors.
• Extensive validation using cross-validation techniques, including leave-one-out and 85/15 split tests.

Main Results:

• All four developed models demonstrated encouraging predictability.
• The ASM-ATC-LOGP model achieved the highest performance with a leave-one-out q2 of 0.832 and RMSE of 0.840 logarithm units.
• Cross-validation tests confirmed the robustness and reliability of the ASM-ATC-LOGP model.

Conclusions:

• The developed aqueous solubility models, particularly ASM-ATC-LOGP, are robust and predictable.
• These models can serve as valuable tools for druglikeness analysis.
• The models offer an efficient method for prioritizing compound libraries before high-throughput screening (HTS).