In the current era of high-throughput sequencing and structure determination, functional annotation has become a bottleneck in biomedical science. Here, we show that automated inference of molecular function using functional linkages among genes increases the accuracy of functional assignments by > or =8% and enriches functional descriptions in > or =34% of top assignments. Furthermore, biochemical literature supports >80% of automated inferences for previously unannotated proteins. These results emphasize the benefit of incorporating functional linkages in protein annotation.