π-PhenoDrug: A Comprehensive Deep Learning-Based Pipeline for Phenotypic Drug Screening in High-Content Analysis

Image-based screening is an efficient drug discovery strategy. Combining high-content screening (HCS) with artificial intelligence efficiently detects drug-induced cellular phenotypes from a large pool of compounds. However, previous studies primarily focus on cell classification and cannot interpret models using morphological features. Additionally, the existing morphology-based studies use only few features, which cannot accurately characterize cell phenotypic perturbations. Herein, π-PhenoDrug, a deep learning-based pipeline, is developed for cell phenotype-driven drug screening. It integrates cell segmentation, morphological profile construction, and phenotype analysis modules of HCS processes. π-PhenoDrug is applied to evaluate drug response in various human melanoma cell lines. The results demonstrate that π-PhenoDrug can evaluate drug killing effects across different cell lines via both supervised and unsupervised modes. Furthermore, π-PhenoDrug identifies drugs with potential killing effects on melanoma cells from a library of diverse compounds. Compared with traditional single-readout assays, this method is more sensitive to compounds that induce weaker phenotypes, reducing the risk of overlooking effective drugs. These results confirm that π-PhenoDrug can achieve high-throughput and accuracy analysis of cell phenotypic data using an unbiased and automated workflow, improving drug discovery efficiency.