Measuring cell state from images has many applications in drug discovery and functional genomics research. The morphology of single cells encodes cellular structure variations that can reveal their function, which makes imaging a powerful source of quantitative information for biological analysis. However, these phenotypic variations are often complex to measure or sometimes cannot be detected by eye (e.g. in multiplexed imaging), requiring efficient computational methods to capture morphological features. In this talk, we will discuss advances in representation learning for images, including weakly supervised learning and contrastive learning, and how these approaches can be used to transform microscopy images of cells into useful biological readouts.