AMALPHI: A Machine Learning Platform for Predicting Drug-Induced PhospholIpidosis.

Area of Science:

• Pharmacology
• Computational Chemistry
• Toxicology

Background:

• Drug-induced phospholipidosis (PLD) is cellular accumulation of phospholipids, often in lysosomes.
• It is linked to cationic amphiphilic drugs (CADs) and can confound drug repurposing, notably for SARS-CoV-2 antivirals.
• Early identification of PLD-inducing compounds is crucial for drug safety and accurate efficacy assessment.

Purpose of the Study:

• To develop predictive machine learning models for identifying potential drug-induced phospholipidosis (PLD) in drug candidates.
• To create a reliable tool for medicinal chemists to assess PLD risk early in the drug development pipeline.

Main Methods:

• A dataset of 545 curated small molecules from ChEMBL v30 was used to train machine learning classifiers.
• Balanced random forest algorithm was employed to build the most effective predictive model.
• Model performance was evaluated using metrics such as AUC (Area Under the Curve).

Main Results:

• The developed machine learning model demonstrated high predictive performance, achieving an AUC of 0.90 in validation.
• The balanced random forest classifier proved to be the most effective in predicting PLD potential.
• The model successfully identified potential PLD inducers among small molecules.

Conclusions:

• Machine learning offers a viable approach for predicting drug-induced phospholipidosis (PLD) potential.
• The developed model and the AMALPHI web platform provide a valuable tool for early-stage drug safety evaluation.
• This predictive capability can significantly aid medicinal chemists in selecting safer drug candidates and avoiding misleading in vitro results.