Prediction of aqueous solubility based on large datasets using several QSPR models utilizing topological structure...

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Discovery

Background:

• Accurate prediction of intrinsic aqueous solubility (S(o)) is crucial for drug development and environmental risk assessment.
• Existing Quantitative Structure-Property Relationship (QSPR) models have limitations in predicting solubility across a wide range of chemical structures.

Purpose of the Study:

• To develop and validate robust QSPR models for predicting intrinsic aqueous solubility of neutral compounds.
• To compare the performance of different modeling techniques, including regression and artificial neural networks.
• To identify key molecular descriptors influencing aqueous solubility.

Main Methods:

• Development of multiple linear regression (MLR), partial least squares (PLS), and artificial neural network (ANN) models.
• Utilized a dataset of 5,964 neutral compounds, categorized into aromatic and non-aromatic classes.
• Employed topological structure descriptors and employed genetic algorithms or backward elimination for descriptor selection.

Main Results:

• All developed models exhibited good performance in external validation, with R² values ranging from 0.72 to 0.84.
• The ANN model, termed CSLogWS, demonstrated superior predictive accuracy compared to eight other existing models.
• E-State and hydrogen E-State descriptors were identified as significant contributors to the ANN model's predictive power.

Conclusions:

• The developed ANN model (CSLogWS) provides a highly accurate and reliable method for predicting intrinsic aqueous solubility.
• This model offers a valuable tool for early-stage drug discovery and environmental safety assessments.
• The study highlights the importance of specific molecular descriptors in QSPR modeling for solubility prediction.