Abstract Prognostic gene signatures are critical in cancer prognosis assessments and their pinpoint treatments. However, their network properties remain unclear. Here, we obtained nine prognostic gene sets including 1439 prognostic genes of different cancers from related publications. Four network centralities were used to examine the network properties of prognostic genes (PG) compared with other gene sets based on the Human Protein Reference Database (HPRD) and String networks. We also proposed three novel network measures for further investigating the network properties of prognostic gene sets (PGS) besides clustering coefficient. The results showed that PG did not occupy key positions in the human protein interaction network and were more similar to essential genes rather than cancer genes. However, PGS had significantly smaller intra-set distance (IAD) and inter-set distance (IED) in comparison with random sets (p-value < 0.001). Moreover, we also found that PGS tended to be distributed within network modules rather than between modules (p-value < 0.01), and the functional intersection of the modules enriched with PGS was closely related to cancer development and progression. Our research reveals the common network properties of cancer prognostic gene signatures in the human protein interactome. We argue that these are biologically meaningful and useful for understanding their molecular mechanism. Keywords: prognostic genes, prognostic genes sets, network property, human protein interactome, cancer, modules 1. Introduction Prognostic genes (PG) have many crucial clinical applications, such as accurate predictions of cancer types (or subtypes), stages and their survival time for cancer patients. In particular, precise targeted treatments and surveillance strategies could be implemented when patients have been classified into different risk groups by means of application of PG [[34]1]. In the past 20 years, there have been tremendous efforts to investigate PG, and a large amount of prognostic gene signatures have been identified in different cancers [[35]2,[36]3,[37]4,[38]5,[39]6,[40]7,[41]8,[42]9,[43]10]. Some PG have been playing important roles in the prognosis of certain cancers, such as ER and HER2 for breast cancer [[44]11]. Biological networks provide a convenient platform of complex relationship studies between biomolecules to trace genetic phenomena and disease mechanisms on a system level [[45]12,[46]13,[47]14]. Network topology analysis helps to discover groups of nodes with special network characteristics in biological networks, as well as associations between groups (e.g., plant immunity [[48]15] and human disease [[49]16,[50]17]). Among them, topological research on cancer genes has showed that they tend to have higher degree and betweenness compared with essential genes [[51]18,[52]19]. Systematic studies of PG’s network properties could help identify pan-cancer PG and unveil their possible mechanisms. However, previous studies of PG were scattered and mostly focus on one specific cancer type or subtype. Cumulative evidence also showed that even for the same cancer type, prognostic gene sets (PGS) obtained by different researchers had very small overlap and questionable reproducibility [[53]20]. Few PG studies have been carried out involving multiple cancer types (e.g., [[54]21,[55]22,[56]23]). Furthermore, they either didn’t pay attention to the network properties of PG or were only involved in topological properties of the PG’s co-expression network properties in a few cancers. Thus, we still know very little about the common topological properties of the human protein interactome. In this study, we first selectively collected 1439 PG of different cancers from 23 related publications and divided them into nine PGS. Based on two protein interaction networks (Human Protein Reference Database and String) and four other gene sets for comparison (cancer gene set: CA, essential gene set: ES, housekeeping gene set: HK, and metastasis-angiogenesis gene set: MA), we then systematically examined their eight network properties including three novel topological measures we proposed. Our study showed that although PG did not possess higher network centralities than CA, PGS had tighter network connections and closer inter-gene set distances than background, and the network modules they were in had many common functions that were closely related to cancer. These findings could help us better understand their roles in complex networks and their mechanisms. 2. Materials and Methods 2.1. Prognostic Genes and Other Four Gene Sets To obtain reliable cancer prognostic gene signatures, we carefully selected 23 related publications from PubMed on the basis of two screening criteria, and each publication had reported one or more cancer signatures that contain 3-300 prognostic genes. Considering the size and type of the cancer signatures, we merged these genes into nine gene sets, each of which consisted of 100 to 200 prognostic genes ([57]Table 1). More details of the selected publications and their screening criteria and the prognostic gene list can be found in [58]Tables S1 and S2. Table 1. List of literature sources, cancer types and sizes of prognostic gene sets in this study. Study ^1 Disease Number of Prognostic Genes in Study Gene Set Number of Prognostic Genes in Gene Set Gentles et al. (Nat. Med. 2015) Multiple tumor types various S1 120 * The Cancer Genome Atlas Research Network (Nature. 2011) Ovarian carcinoma 190 S2 185 Lenz et al. (N. Engl. J. Med. 2008) (Diffuse) Large-B-cell lymphomas 39,283,71 S3 330 Zhao et al. (PLoS Med. 2006) Renal cell carcinoma 259 S4 222 Dave et al. (N. Engl. J. Med. 2006) Burkitt’s lymphoma 217 S5 200 Bullinger et al. (N. Engl. J. Med. 2004) Acute myeloid leukemia (AML) 133 S6 103 Liu et al. (J. Natl. Cancer Inst. 2014) (Triple-negative) Breast cancer 11 S7 135 Wang et al. (Lancet. 2005) (Lymph-node-negative) Breast cancer 76 van de Vijveret al. (N. Engl. J. Med. 2002) Breast cancer 70 Wistuba et al. (Clin. Cancer Res. 2013) Lung adenocarcinoma 31 S8 118 Tang et al. (Clin. Cancer Res. 2013) Non-small cell lung cancer (NSCLC) 12 Xie et al. (Clin. Cancer Res. 2011) NSCLC 59 Zhu et al. (J. Clin. Oncol. 2010) NSCLC 15 Boutros et al. (Proc. Natl. Acad. Sci. USA 2009) NSCLC 6 Lau et al. (J. Clin. Oncol. 2007) NSCLC 3 Gerami et al. (Clin. Cancer Res. 2015) Melanoma 28 S9 174 Wu et al. (Proc. Natl. Acad. Sci. USA 2013) Prostate cancer 32 Li et al. (J. Clin. Oncol. 2013) AML 24 Lohavanichbutr et al. (Clin. Cancer Res. 2013) Oral squamous cell carcinomas (OSCC) 13 Sveen et al. (Clin. Cancer Res. 2012) Colorectal cancer 7 Smith et al. (Gastroenterology. 2010) Colon cancer 34 Ramaswamy et al. (Nat. Genet. 2003) Solid tumors 17 Yeoh et al. (Cancer Cell. 2002) Acute lymphoblastic leukemia (ALL) 7–20 [59]Open in a new tab ^1: Please see [60]supplementary Table S1 for details of references; *: