We obtained multidimensional genomics data across 32 cancer types from the TCGA consortium, including mutation, copy number and expression data. The mutation and copy number data for cell lines are obtained from the CCLE and GDSC resource.

(1) Mutation, copy number alterations and expression profiles are derived across 32 cancer types from TCGA consortium, including 8580 patient samples.

(2) ShRNA data were downloaded from The Project Achilles database, which provides shRNA depletion scores from pooled genomic library tests across 216 cancer cell lines.

(3) High-throughput CRISPR-Cas9 screening data, including 63 cancer cell lines, were downloaded from the GenomeCRISPR database.

(4) A genome-scale genetic interaction map in yeast was obtained from Costanzo et al. (Costanzo M, Baryshnikova A, Bellay J et al. The genetic landscape of a cell. Science 2010; 327: 425-31.).

(5) The original pharmacological screening data were downloaded from Cancer Cell Line Encyclopedia (CCLE), Genomics of Drug Sensitivity in Cancer (GDSC), Broad Cancer Therapeutics Response Portal (CTRP, http://www.broadinstitute.org/ctrp/) and NCI60.

(6) Synthetic viable or synthetic lethal interactions were integrated from 17 studies (Table 1)

In the Search page, users can key in gene ID or symbol to perform a search of corresponding genetic interactions (1), which can be further screened by advance options (2-3).

When user queries the genes to the database, the CGIdb will provide search results on this page. The search results are divided into two main parts: (i) information box, which includes the basic information of your selected gene, and the external link to NCBI for more detail (1); (ii) The SL/SV pairs list is provided on the bottom of page (2). And user can click the button (3) to view the details of the gene pairs, including drugs effect and protein-protein interaction network. The results can be exported as CSV format (4).

The details of results are divided in three parts. (i) Distribution of gene alteration in TCGA data are provided. CGIdb shows the altered samples (mutation and copy number alteration) of genes and statistic of analysis results (1-2). (ii) In the middle of page, user can get results of the drug effect. In cell lines with drug targeting the searching gene, a one-sided Wilcoxon rank sum test is used to test whether the drug response measures, such as IC50, are significantly higher or lower in cells lines with and without alterations of the partner genes in the genetic interaction (3). Detailed information about the pharmacodynamic data are shown in the right table (4-5). (iii) On the bottom of the detail page, CGIdb provides the visualized network of protein-protein interactions derived from PathwayCommon. User can export network as jpg or png file (6-7).

In the Browse page, users can easily filter SV and SL pairs classified by tissue types. The human body on the left includes optional tissue types (1). Click on the tissue on the model to filter the genetic interactions of the tissue. The detailed results are showed in the right table (2).

In the data page, users can download or upload data, We provide all SV and SL pairs, which are classified by different sources and tissue types (1-3).

The SV and SL interactions analyzed in the present study were obtained from different types of sources, including computational predictions, biochemical assays, and text mining results. In addition, biochemical assays were based on different experimental technologies and platforms, such as shRNA, CRISPR and drug inhibition. Because multiple types of evidence are conducive to the identification of SV (SL) interactions, an integrative confidence score combining scores from these evidence sources can provide an overall estimation of the reliability of a SV (SL) interaction. In principle, we supposed that (i) the contribution of experimental evidence to the confidence score is more significant than the contribution of predictive algorithms or text mining and that (ii) the SV (SL) interactions supported by more evidence sources should be beneficial to the confidence score. The scoring procedures were divided into two steps, i.e., quantification and integration. A large number of SV (SL) interactions collected from other studies had only qualitative annotation evidence (such as “high-throughput” or “low-throughput”), or technological descriptions of wet-lab experiments (such as “CRISPR screening” or “shRNA screening”). Thus, it was necessary to assign quantitative scores to these SV (SL) interactions before the calculation of integrative scores. Similar to the scoring scheme from SLDB, the quantitative scores were assigned based on the experimental methods as shown in the following Table 2. For instance, “Mutant & Mutant” indicated that the pair of SV (SL) genes were disturbed by transgenic or genetic deletions. Moreover, “RNA interference & Mutant” indicated that one gene was perturbed by RNAi and that the other was perturbed via mutation. In summary, the SV (SL) interactions obtained from low-throughput experiments were considered to be more reliable than the results from high-throughput experiments due to the lower false positive rate. However, a higher confidence score was assigned to low-throughput evidence than high-throughput evidence. Compared to other RNA interference experiments (such as shRNA, siRNA and dsRNA), the CRISPR screen had lower off-target effects, which were assigned higher confidence scores similar to mutation and transfection experiments (Table 2).

The formula to combine the individual scores as follows：

where s represents the integrative score corresponding to the experimental evidence; pi is the individual score; and n is the total number of experimental supporting evidence.