As the largest natural source of sulfur-containing gases into the atmosphere, ocean organism-derived dimethyl sulfide (DMS) has been considered to play a critical role in the Earth’s climate system. Yet there are great uncertainties in modeling the spatiotemporal variations of DMS and incomplete knowledge of influencing factors in different oceanic regions. Moreover, little is known about the future change of global DMS, which limits our understanding of the feedback of marine ecosystem to climate change. Here we develop an artificial neural network model and combine data mining approaches to address these issues. Phytoplankton biomass and salinity are currently predominant factors associated with DMS variability in the coastal and Arctic regions, respectively. In the mid- and low-latitude open oceans, nutrients and temperature are also crucial factors in addition to radiation and mixed layer depth, and their relationships with DMS show reversals when passing certain thresholds. Although the global average DMS concentration and emission slightly decline from 2005 to 2100, they may change considerably in specific regions. In contrast to the DMS decreases in the low-latitudes mainly related with phosphate reduction and temperature rise and in the North Atlantic subpolar gyre attributed to salinity decline, warming will cause DMS increase in the Southern Ocean and sea ice loss will dramatically enhance DMS emission in the Arctic. Although the global negative feedback loop between oceanic DMS and climate may not operate, the future spatial redistribution of DMS may lead to the change in cloud cover pattern and significantly affect regional climate.