Deciphering the complexity of spatial-temporal proteomics via subcellular location optimal transport

The localization of intracellular proteins is highly dynamic, which segregates protein expression and plays a crucial role in the functional polarization of cells. Recent advancements in technology have facilitated the high-throughput measurement of subcellular spatially resolved protein expression. However, methods to analyse these data are not well established. We present a machine-learning-based approach called Subcellular Location Optimal Transport (SLOT) to decipher the complexity of spatial-temporally variable proteins. By comparing the spatial expression patterns of different proteins, SLOT identifies potential spatially co-localized proteins, shedding light on their functional relationships within the cell. SLOT graphs are constructed for proteins at different time points, and these temporal SLOT graphs are aligned across time series to elucidate dynamic changes in protein co-localization patterns over time. We applied SLOT to reveal spatial-temporal proteomic changes in Xenopus oocytes and early embryos, uncovering previously unexplored intricacies in protein localization dynamics during early developmental stages. These findings underscore the utility of SLOT in providing mechanistic insights into fundamental cellular processes, such as development and differentiation. Our study demonstrates that SLOT is a powerful analytical tool for deciphering the spatial-temporal dynamics of protein localization within cells. Furthermore, the insights gained from SLOT analysis enhance our understanding of key cellular processes, offering new avenues for research in developmental biology and beyond.