A Model-Based Clustering Approach for Chemical Toxicity Assessment Using Cell Painting Data

Risk assessment of chemicals relies heavily on toxicity studies conducted on laboratory animals. However, those studies are lengthy, costly, and are not always fully relevant to human physiology. This is where human-derived, cell-based assays, such as the Cell Painting technique, could play a crucial role. This high-content imaging approach allows for the analysis of several cellular compartments, offering a comprehensive view of how these chemicals impact cell morphology and organelle structure.Our goal in this study is to simplify complex dose-response analysis in high dimensions by using an intermediate clustering model of the cells within each well of the assay plates. Specifically, we focus on examining the wells to analyze and compare the clusters obtained from cells treated with chemical compounds at different concentrations.To achieve this, we propose a model-based clustering approach for toxicity assessment using well-level data from Cell Painting experiments. This data consists of measurements taken within each well, where chemical compounds are tested at different concentrations (8 in our case). We assume that each well contains different classes of cells. We develop a mixture model to describe the distribution of cell characteristics within wells, allowing class proportions to depend on compound concentration. We also assume that the distribution of features across different clusters is the same. In this approach, we use a Gaussian Mixture Model. The parameters of the model are estimated using the EM algorithm.In the numerical experiments, we apply our algorithm to real data from Cell Painting experiments to assess its effectiveness. The goal is to evaluate how well the model identifies shifts in class proportions across different compound concentrations. Finally, this approach enables the identification of consistent patterns across experimental conditions and provides insights into how varying concentrations affect cell behavior. The model can also be extended to account for different cell lines.