Cancer is the second disease leading to death worldwide [1]. It consists of more than 100 different diseases with diverse risk factors. Among these risk factors, genetic alternations play critical roles in the pathogenesis of the disease and provide fundamental clues for the identification of drug targets and the development of novel drugs [2–6]. Recently, several large-scale cancer genome projects produced multi-dimensional genome-wide big data such as The Cancer Genome Atlas (TCGA) [7], the Wellcome Trust Sanger Institute's Cancer Genome Project [8], and the International Cancer Genome Consortium (ICGC) [9]. These genome-wide data have dramatically advanced cancer research, especially in terms of its genetics and genomics [10], which enhances the accuracy and coverage of the identification of cancer-related genes that could drive or protect cancer development. However, though these large-scale sequencing data discovered thousands of mutations, they have not so far identified novel drug targets besides these previously identified [11]. Therefore, developing novel approaches to revealing the signal buried under the big data to discover the novel drug targets is necessary and critical for the development of effective treatment for cancer.

Among these cancer-related genes identified by high-throughput sequencing and small-scale traditional approaches, two classes of genes - tumor suppressor genes (TSGs) and oncogenes (OCGs) - have been attracted much attention. Numerous studies demonstrated these genetic alternations involve the gain-of-function of OCGs together with the loss-of-function of TSGs determine the cell cycle processes that control the tumor formation and development [12, 13]. Recently, protein-protein interaction (PPI) network based on computational methods have been used to identify disease-specific genes, modules, and cancer-subtype subnetworks [14–16]. Therefore, we hypothesized that comparative investigations of TSG and OCG mutation patterns and network properties would provide a number of novel insights into their functions in the tumorigenesis, which further offers valuable information for identification of novel drug targets for drug development.

As numerous genetic and genomic data in cancer become available, the list of OCGs and TSGs has been expanded through molecular, cellular, genomic, and computational studies including non-coding RNA genes [17, 18]. Considering the gain-of-function of OCG mutations and loss-of-function of TSG mutations, TSGs and OCGs may be involved in the regulation of cellular functions in a yin-yang fashion [19]. For example, our previous study has shown that they have distinct and competitive regulatory patterns in ovarian cancer [20]. Furthermore, OCG mutations are usually dominant so that one mutant copy is enough to start switching on a cellular activity. TSG mutations tend to be recessive, so that they should follow the famous Knudson's 'two-hit hypothesis': that both copies of tumor suppressor genes need mutate to cause loss of function. However, more and more evidence shows that even partial inactivation of TSGs could critically contribute to tumorigenesis [21]. Additionally, some genes' function could be switched between OCGs and TSGs, depending on the situation. Current therapeutic applications have shown that targeting OCGs and their related pathways is promising for developing novel drugs, including antibodies and small synthetic molecules [22]. Therefore, further understanding of OCGs and TSGs in the terms of networks will provide novel insights into the their functions in the tumorigenesis. However, to our knowledge, there is no report that systematically investigates their relationships.

Thus, in this study, we compared five sets of proteins encoded by five sets of genes (TSGs, OCGs, drug target genes, essential genes, and other genes) with the perspectives of genomics and protein networks. We compared them using the somatic mutations from TCGA Pan-Cancer project [23] and network properties in human PPI networks [24]. Based on the genetic data from Pan-Cancer project, we found that TSGs had the highest mutation frequency in most tumor types and the OCGs second. For the network properties, relative to target proteins, essential proteins, and other proteins, both TSG and OCG proteins tended to have higher degrees, higher betweenness, lower clustering coefficients, and shorter shortest-path distances. In addition, both TSG and OCG proteins tended to have direct interactions with cancer drug target proteins. We further generated a TSG-OCG network and found that TSGs and OCGs connected strongly with each other. Our study first revealed that the OCGs and TSGs had different mutation patterns, but had similar and stronger protein-protein characteristics relative to the essential proteins or control proteins in the whole human interactome.

Cancer is a genetically complex disease, which involves the combined functions of tumor suppressor genes (TSGs) and oncogenes (OCGs). TSGs and OCGs jointly play important roles in the cancer development through loss of function or gain of function. Most of them cannot trigger the cancer development by themselves. Numerous studies about genetic alterations of TSGs and OCGs, especially OCGs, have led to the identification of drug targets for cancer treatment. However, the identification of novel drug targets has become more challenging even though genome-wide sequencing data provide thousands of mutations. Therefore, development of novel approaches for identification of novel drug target is mandatory. To facilitate the development of novel approaches, in this study, we comprehensively compared TSGs and OCGs from the perspectives of somatic mutation and network properties. These broad comparative results allow us to address several questions that might be useful for the development of new methods: 1) Do TSGs and OCGs have similar or different mutation frequency patterns? 2) How do they relate to each other? 3) How do they relate to cancer drug target? 4) Do the TSGs and OCGs tend to link closely to each other? The results indicated that while the TSGs and OCGs had different mutation patterns, they had similar network characteristics. They were also not only related to each other closely, but also to cancer drug targets.

In this study, we mainly focused on the examination of the mutation patterns of TSGs and OCGs from the whole-genome wide data in the Pan-Cancer project [23]. It was different from the purpose of the Pan-Cancer analysis project. The study of Pan-Cancer analysis presents the data and analytical results for point mutations and small insertions/deletions from 3,281 tumours across 12 tumour types as part of the TCGA Pan-Cancer effort. They illustrated the distributions of mutation frequencies, types and contexts across tumour types, and establish their links to tissues of origin, environmental/carcinogen influences, and DNA repair defects. However, they did not go further to examine the mutation patterns of TSGs and OCGs. In this study, we separated the mutation data of TSGs and OCGs from the rest genes and performed a comparison of five gene sets. We found that the TSGs had the highest mutation frequency in most tumour types and the OCGs second. The results might be interpreted by the theory that the gain-of-function mutations that convert proto-oncogenes to oncogenes acts dominantly while the loss-of-function mutation in tumor suppressor genes acts recessively. In addition, we observed that the essential genes had the lowest mutation frequency in all tumor types, which might reflect the fundamental roles in the survival of the essential genes.

However, we did not dive further to study the consequence of or causal relation to mutations for the function roles of TSGs and OCGs. As we known that TSGs and OCGs have different roles during the cancer development. However, it is not very clear how they work together. It will be very interesting and useful to further study the association between the mutation frequency and the roles of TSGs and OCGs. For example, we observed that the mutation frequency of TSGs was about two times of that of OCGs. It is not clear whether or not this mutation frequency difference influence or linked to their functional roles in the pathogenesis of cancer. Moreover, it is very challenging to assess the association between the mutation frequency difference and functional roles of TSGs and OCGs by both computational and experimental examination.

In this study, we compared the drug target genes with TSGs and OCGs in the view of mutation frequencies and network properties. We found drug target genes generally tend to have less mutations compared to TSGs and OCGs and also have lower degrees. These results suggested that the genetic contribution of drug target genes is not strong as TSGs and OCGs. Besides, we found both TSG and OCG proteins tended to have direct interactions with cancer drug target proteins. However, we did not further examine if the drug targets either suppress actions on oncogene activity or restore TSG functions through direct interaction or indirectly interactions. It might be very interesting to further examine if the mutations in OCGs or TSGs are necessary for both the establishment and maintenance of protein-protein interactions, which might lead to the identification of logical drug targets. However, to map the mutation to proteins for detecting the mutation-specific perturbations at the network level need much efforts including the development of protein structure-guided pipeline for extracting interacting protein sets specific to a particular mutation, which is beyond of the scape of this study. In the future, we will integrate the protein structure information with mutation information in the context of PPI network to further understand the connection of TSG and OCG proteins in the cancer development.

The study was mainly based on the data coming from both public data and predicted results. As most of the computational biology studies, it is very challenging to obtain the error-free or complete data. Therefore, in the analysis process, there still have several steps we could improve in the future, including the selection of gene sets, specification of protein function association data, and mutation data of cancer with less bias. For gene sets, we chose the genes with high confidence for analysis. The data set used here are far from complete and error-free. For the protein associations, we utilized the PPI data from PINA database, which includes the physical association, genetic association, and enzymatic reaction curated from six other databases. It is not clear about how these mutations alter the interaction relationship with their partners. For mutation data from cancer patients, we mainly utilized the data from TCGA, which might be biased by sequencing depth, platform, and sample size. However, our analysis still provided statistically significant characteristics of somatic mutations and networks of TSGs and OCGs. The list of TSGs and OCGs is updated frequently based on different methods. Therefore, the characteristics of TSGs and OCGs under investigation will not be exactly the same as those we concluded here. However, the tendencies we obtained in this study might provide some clues for further investigation of functional roles of TSGs and OCGs in carcinogenesis and identification of novel drug targets.

In this study, we explored the somatic mutation and network characteristics of TSGs and OCGs. Based on the mutation data from Pan-Cancer project, we found that the TSGs had the highest mutation frequency. Based on the human protein-protein interaction network, we found that TSG proteins and OCG proteins had similar global network topological characteristics and that the TSGs, OCGs, and drug targets had a tendency to interact with each other. Integration of mutation frequency with TSG-OCG network provided insights that TSGs and OCGs might jointly contribute to the cancer development. In summary, this study first comprehensively investigated TSGs and OCGs from the perspective of genetics and networks, which provides novel insight into the roles of TSGs and OCGs in cancer development and treatment.