Results and Discussion

Database and web interface
Based on the reported cancer-related gene sets from CGC and CGI, a cancer gene database and a cancer pathway gene database were constructed by the described methods. In total, 436 CGC-originated and 7181 CGI-originated cancer genes were prepared for the reported cancer gene annotation. Additionally, 5,790 CGC-based and 6,744 CGI-based cancer pathway genes were assigned for the annotation of unreported but potential candidate genes (Table 1). The gene ID database for gene symbol conversion, based on HGNC, and the functional annotation database, based on Entrez Gene and UniProt database flat files, were constructed. The types of annotations provided by CaGe are summarized in Table 2.
The main window of the CaGe web interface is shown in Fig. 2 for the cancer gene annotation process. Main menus, located in the upper part of the main window, are linked to the four main functions and home page of the system: 1) home page, 2) cancer gene annotation function, 3) cancer pathway annotation function, 4) cancer gene browsing function, and 5) cancer pathway browsing function. Detailed usages for CaGe are described in the user's guide at http://mgrc.kribb.re.kr/cage/include/cageUserGuide.pdf.

System information
CaGe was developed using PHP, R, and python languages; MySQL for database management; and Apache for the web server and is operated on a Linux platform with 8-core Intel Xeon CPUs (2.33 GHz) and 24 GB of main memory.

Performance evaluation with small cell lung cancer mutation data
To assess the capability of CaGe, we annotated genes from candidate mutations for a small cell lung cancer genome (SCLC) [12]. We had 59 genes with predicted functionally damaging mutations after applying 22,910 mutations to the PolyPhen and could annotate 22 previously reported known cancer genes successfully by applying the PolyPhen output for CaGe. Known cancer genes included RB1, which was mentioned as an SCLC-related gene in Pleasance's work; DST and ETS2, which were previously reported as SCLC-related genes but not mentioned in Pleasance's work; and 3 more genes (PDGFC, IL16, and AGTR2), which are known as lung cancer-related genes (Supplementary Table 1). The other 16 genes are known to be related to other cancer types, suggesting that they might be important in the carcinogenesis of SCLC as well. From the pathway analysis, we identified 12 known cancer genes and 11 genes in cancer-related pathways (Supplementary Table 2). Those 11 genes might also be important in the carcinogenesis of SCLC. Thus, we conclude that CaGe can annotate cancer genes effectively and suggest that CaGe will be useful in the identification of cancer-causing mutations and genes in HGT-based cancer genomics.
In this paper, we present a new cancer genomic tool, CaGe, for the assessment of candidate cancer genes having somatic mutations from HGT-based cancer genomics. CaGe provides users with information on cancer genes, mutations, pathways, and associated annotations through cancer gene annotation, cancer pathway annotation, cancer gene browsing, and cancer pathway browsing functions. It has a capacity to process SIFT or PolyPhen output files as direct input for usual NGS-based cancer genomics flows. Researchers can classify their candidate genes from cancer genome studies into previously known or novel categories of cancer genes and gain insight into the underlying carcinogenic mechanisms through a pathway analysis. We hope that CaGe will be useful for the identification of cancer-causing mutations and genes in HGT-based cancer genomics.