This paper explores the scalability of weakly supervised classifiers and self-supervised masked autoencoders (MAEs) in the context of microscopy image analysis for biological research. The authors investigate the performance of these models on large-scale datasets, including the RxRx3 and JUMP-CP datasets, which contain millions of images from various genetic and chemical perturbations. They find that ViT-based MAEs outperform weakly supervised classifiers, achieving up to an 11.5% relative improvement in recalling known biological relationships curated from public databases. Additionally, they introduce a new channel-agnostic architecture (CA-MAE) that allows for inputting images with different numbers and orders of channels at inference time, demonstrating effective generalization to microscopy datasets with different channel configurations. The findings suggest that self-supervised learning on microscopy data can lead to powerful foundation models of cellular biology, with potential applications in drug discovery and other areas. The relevant code and models are available on GitHub.This paper explores the scalability of weakly supervised classifiers and self-supervised masked autoencoders (MAEs) in the context of microscopy image analysis for biological research. The authors investigate the performance of these models on large-scale datasets, including the RxRx3 and JUMP-CP datasets, which contain millions of images from various genetic and chemical perturbations. They find that ViT-based MAEs outperform weakly supervised classifiers, achieving up to an 11.5% relative improvement in recalling known biological relationships curated from public databases. Additionally, they introduce a new channel-agnostic architecture (CA-MAE) that allows for inputting images with different numbers and orders of channels at inference time, demonstrating effective generalization to microscopy datasets with different channel configurations. The findings suggest that self-supervised learning on microscopy data can lead to powerful foundation models of cellular biology, with potential applications in drug discovery and other areas. The relevant code and models are available on GitHub.