Interdisciplinary Study on Drug-Induced-Phospholipidosis of Repurposing Libraries through Machine Learning and Experimental Evaluation in Different Cell Lines

Phospholipidosis is a cellular condition characterized by the excessive accumulation of phospholipids within cells, that also can be induced by medications—especially those known as cationic amphiphilic drugs. While this phenomenon is not always harmful in itself, it can interfere with how cells respond in laboratory experiments and complicate the interpretation of drug screening results, leading to potential delays or failures in the development of new therapies. As the pharmaceutical industry increasingly relies on high-throughput screening and machine learning to accelerate early-stage drug discovery, there is growing interest in identifying compounds that may cause phospholipidosis before they move too far through the development pipeline. In response to this challenge, we at the EU-Horizon HLTH Remedi4ALL project compiled and analyzed one of the most comprehensive datasets to date—over 5,000 repurposed drugs tested across multiple cell lines—to better understand which compounds are likely to induce phospholipidosis. Using this dataset, we developed a machine learning model capable of predicting the risk of phospholipidosis based on a drug’s chemical structure. By validating our approach across diverse experimental settings, our goal is to provide the broader scientific and pharmaceutical communities with a practical tool to flag problematic compounds early, ultimately helping bring safer, more effective treatments to patients faster. – A comprehensive dataset for a repurposing collection, comprising 5,228 compounds, for induction of phospholipidosis in A549-ACE2 and Vero-E6 cells –A machine learning model developed using in-house experimental phospholipidosis data across multiple cell lines –Exploration of chemical features that induce phospholipidosis and investigation of the relationships underlying different cell line susceptibilities –A case study: correlation of drug-induced phospholipidosis with antiviral compound effects against SARS-CoV-2 using phenotypic screening data Phospholipidosis (PLD), a cellular adverse effect that is, among others, caused by numerous cationic amphiphilic drugs. Interest is raised within pharma discovery to predict this phenomenon, as it can impact the outcome of phenotypic cellular screens and significantly delay drug development processes. The development of accurate and validated machine learning models for predicting drug-induced PLD across different cell lines and research centers could provide a valuable early application tool for the pharmaceutical industry, potentially accelerating drug discovery and reducing the risk of late-stage failures. We report here the assembly, curation, testing and modeling of one of the largest datasets of repurposed drugs (5K+) tested for PLD induction on different cell lines. A machine-learning classification method was developed and validated to predict whether molecules are prone to induce PLD effects when applied in cell-based screens.