Can generative AI replace immunofluorescent staining processes? A comparison study of synthetically generated cellpainting images from brightfield

Xiaodan Xing, Siofra Murdoch, Chunling Tang, Giorgos Papanastasiou, Jan Oscar Cross-Zamirski, Yunzhe Guo, Xianglu Xiao, Carola-Bibiane Schönlieb, Yinhai Wang, Guang Yang

Abstract: Highlights•This study benchmarks five models across CNN, GAN, and diffusion backbones using a public dataset for a comparative analysis.•Deep learning-generated synthetic immunofluorescence images show high visual quality and can distinguish single-cell mechanisms under various treatments.•Model generalizability is a concern, with performance dropping on unseen cell lines, raising trustworthiness issues.•The generalisability of the synthesis models is a concern. When tested on unseen cell lines, performance dropped sharply. Additionally, despite high mechanisms of action prediction accuracy, most feature correlations are low, raising concerns about the trustworthiness of these algorithms.