Motivation Essential gene signatures for cancer growth have been typically identified via RNAi or CRISPR-Cas9. Here, we propose an alternative method that reveals the essential gene signatures by analysing genomic expression profiles in compound-treated cells. With a large amount of the existing compound-induced data, essential gene signatures at genomic scale are efficiently characterized without technical challenges in the previous techniques. Results An essential gene is characterized as a gene presenting positive correlation between its down-regulation and cell growth inhibition induced by diverse compounds, which were collected from LINCS and CGP. Among 12741 genes, 1092, 1228827962, 1664580 and 829 essential genes are characterized for each of A375, A549, BT20, LNCAP, MCF7, MDAMB231 and PC3 cell lines (P-value 1.0E-05). Comparisons to the previously identified essential genes yield significant overlaps in A375 and A549 (P-value 5.0E-05) and the 103 common essential genes are enriched in crucial processes for cancer growth. In most comparisons in A375, MCF7, BT20 and A549, the characterized essential genes yield more essential characteristics than those of the previous techniques, i.e. high gene expression, high degrees of protein-protein interactions, many homologs and few paralogs. Remarkably, the essential genes commonly characterized by both the previous and proposed techniques show more significant essential characteristics than those solely relied on the previous techniques. We expect that this work provides new aspects in essential gene signatures. Availability and implementation The Python implementations are available at https://github.com/jmjung83/deconvolution_of_essential_gene_signitures. Supplementary information Supplementary data are available at Bioinformatics online.