With the rapid advancements in spatial multi-omics technologies, it has become feasible to acquire and analyse multiple omics data from the single tissue. Investigating the potential of integrating spatial multi-omics data is of great importance for advancing biological research. However, because of technical limitations and high costs, spatial proteomics data remain scarce in comparison to relatively abundant spatial transcriptomics data. Overcoming the paucity of spatial proteomics data could markedly expedite the progress of spatial multi-omics integration, potentially leading to transformative breakthroughs in life science. To this end, we propose STProtein, a novel framework leveraging graph neural networks with multi-task learning strategy to predict unknown spatial protein expression from previous spatial multi-omics data. STProtein is also a robust computational toolkit that enables scientists to boost the scientific discovery like identifying complex hidden protein spatial patterns within tissue; observing undetectable relationship between different maker genes, and even finding "Dark Matter" in unknown world.