Manual annotation for the significant genes from the routine bioinformatics analyses of large-scale omics datasets has become the bottleneck for further biomedical knowledge discovery. To address this challenge, a novel computational strategy called FORECAST was developed to realize seamless fusion between omics datasets and literature information for automatic gene-centric knowledge mining. In the case of cancer gene discovery, FORECAST exhibits superior efficacy over routine strategies of either omics data analysis or literature mining. Identified novel cancer genes also proved the high capability of FORECAST to mine novel knowledge. To meet the requirements of large-scale omics datasets analysis, FORECAST shows great robustness and efficiency when implemented in high-volume, noisy and incomplete biological omics and literature datasets. This strategy is promising to accelerate the knowledge mining in the era of Big Data.