Accurate annotation of gene function in individual samples and even in each cell type is essential for understanding the pathogenesis of cancers. Single-cell RNA-sequencing (scRNA-seq) provides unprecedented resolution to decipher gene function. In order to explore how scRNA-seq contributes to the understanding of gene function in cancers, we constructed an assessment framework based on co-expression network and neighbor-voting method using 116,814 cells. Compared with bulk transcriptome, scRNA-seq recalled more experimentally verified gene functions. Surprisingly, scRNA-seq revealed cell-type-specific functions, especially in immune cells, whose expression profile recalled immune-related functions that were not discovered in cancer cells. Furthermore, scRNA-seq discovered sample-specific functions, highlighting that it provided sample-specific information. We also explored factors affecting the performance of gene function prediction. We found that 500 or more cells should be considered in the prediction with scRNA-seq, and that scRNA-seq datasets generated from 10x Genomics platform had a better performance than those from Smart-seq2. Collectively, we compared the prediction performance of bulk data and scRNA-seq data from multiple perspectives, revealing the irreplaceable role of single-cell sequencing in decoding the biological progresses in which the gene involved.