Development of a phospholipidosis database and predictive quantitative structure-activity relationship (QSAR) models

Area of Science:

• Toxicology
• Computational Chemistry
• Drug Development

Background:

• Drug-induced phospholipidosis (PL) is a lysosomal accumulation of phospholipids and drugs, often seen in preclinical studies.
• PL can hinder or halt pharmaceutical development, necessitating predictive tools.

Purpose of the Study:

• To construct a comprehensive database of PL findings across various animal species.
• To utilize this database for training computational toxicology software (QSAR models) for predicting PL.

Main Methods:

• Compiled PL data and chemical structures from literature, databases, and FDA reports for 583 compounds.
• Developed Quantitative Structure-Activity Relationship (QSAR) models using MC4PC and MDL-QSAR software.
• Assessed model performance via internal cross-validation, focusing on specificity and sensitivity.

Main Results:

• The database included 190 positive and 393 negative PL findings.
• Individual QSAR models showed varying predictive performance (e.g., MC4PC: 78% concordance, MDL-QSAR: 79% concordance).
• Combining models significantly improved predictive performance, achieving 81% sensitivity and 79% specificity.

Conclusions:

• QSAR models, trained on a curated PL database, can effectively predict potential drug-induced phospholipidosis.
• Structural alerts identified by QSAR correlate with cationic amphiphilic drug (CAD) properties, supporting chemical structure-PL relationships.
• These QSAR models offer a valuable tool for early screening of drug candidates to mitigate PL risks in pharmaceutical development.