pHSol

pHSol predicts the pH-dependent aqueous solubility of druglike molecules to support assessment of absorption, distribution, metabolism, excretion (ADME) and toxicity.

Key Features:

• Henderson-Hasselbalch Equation Utilization: Models pH-dependent solubility changes using the Henderson-Hasselbalch equation to account for drug ionization states.
• Artificial Neural Networks (ANNs): Predicts intrinsic aqueous solubility with ANNs trained on a subset of 4,548 druglike compounds from the PHYSPROP database.
• Marvin pKa Integration: Uses the Marvin pKa plug-in to predict acid/base dissociation coefficients (pKa) for determining ionization states.
• Validation and Performance Metrics: The best ANN model achieves a cross-validated RMSE of 0.70 log S-units, Marvin's pKa predictions show an RMSE of 0.71 pH-units, and the combined pH-dependent model attains a mean RMSE of 0.79 log S-units against 27 drugs with experimental pH-solubility curves.
• Application to Large Vendor Libraries: Profiles solubility space at low pH for five large vendor libraries.

Scientific Applications:

• Drug Discovery and Development: Informs formulation strategies and bioavailability optimization by predicting pH-dependent dissolution of druglike compounds.
• ADME/Toxicity Profiling: Provides solubility profiles that support prediction of pharmacokinetic behavior relevant to absorption, distribution, metabolism, excretion (ADME) and toxicity.
• High-throughput Screening: Enables solubility profiling of large compound libraries for early-stage screening in drug discovery.

Methodology:

Implements the Henderson-Hasselbalch equation; employs ANN models trained on a subset of 4,548 PHYSPROP compounds; predicts pKa using the Marvin pKa plug-in validated on 467 PHYSPROP molecules; and validates the combined pH-dependent model against pH-solubility curves for 27 drugs.