Enhancing Cellular Clustering in Malaria Drug Discovery via Unsupervised Learning

Malaria remains a major global health threat, driving the need for novel antimalarial drug discovery techniques. High-content fluorescence microscopy, combined with artificial intelligence (AI), enables large-scale phenotypic screening of Plasmodium falciparum gametocytes. The PHIDDLI (Phenotype-based High-content Imaging for Drug Discovery using Live Imaging) pipeline is one such AI-powered system that automates the analysis of cellular responses to chemical compounds through feature extraction and clustering. However, PHIDDLI currently uses Principal Component Analysis (PCA) and KMeans clustering, which may limit the interpretability of complex phenotypic patterns. In this study, we enhance the pipeline by introducing Kernel PCA and t-distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction, and Hierarchical clustering for phenotypic grouping. Using feature embeddings from fluorescence microscopy images, we evaluate each combination based on biological relevance and visual clarity. Our results demonstrate that t-SNE with Hierarchical clustering produces more distinct and interpretable phenotype groupings, aligning closely with expert biological interpretation. We also present interactive visualizations to support exploratory analysis. These enhancements contribute to more explainable AI in antimalarial research, facilitating faster and more reliable identification of promising drug candidates.