Identification of Hepatocellular Carcinoma–Related Genes with a Machine Learning and Network Analysis

Abstract

Liver cancer is one of the leading causes of cancer mortality worldwide. Hepatocellular carcinoma (HCC) is the main type of liver cancer. We applied a machine learning approach with maximum-relevance–minimum-redundancy (mRMR) algorithm followed by incremental feature selection (IFS) to a set of microarray data generated from 43 tumor and 52 nontumor samples. With the machine learning approach, we identified 117 gene probes that could optimally separate tumor and nontumor samples. These genes not only include known HCC-relevant genes such as MT1X, BMI1, and CAP2, but also include cancer genes that were not found previously to be closely related to HCC, such as TACSTD2. Then, we constructed a molecular interaction network based on the protein–protein interaction (PPI) data from the STRING database and identified 187 genes on the shortest paths among the genes identified with the machine learning approach. Network analysis reveals new potential roles of ubiquitin C in the pathogenesis of HCC. Based on gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis, we showed that the identified subnetwork is significantly enriched in biological processes related to cell death. These results bring new insights of understanding the process of HCC.

1. Introduction

Liver cancer is one of the major cancer tumor types diagnosed worldwide. In males, liver cancer is the fifth most frequently diagnosed cancer and the second most frequent cause of cancer-related death. In females, it is the seventh most commonly diagnosed cancer and the sixth leading cause of cancer-related death (Jemal et al., 2011). Hepatocellular carcinomas (HCC) represent the major histologic type and account for between 85% and 90% of primary liver cancers (El-Serag and Rudolph, 2007). HCC is the fifth most common cause of cancer, and its incidence is still increasing (Llovet et al., 2003). Therefore, it is of great importance to identify HCC-related genes to understand the pathological process.

Maximum relevance–minimum redundancy (mRMR) is an effective machine learning approach for phenotype classification based on transcriptome data (e.g., gene expression microarray), which was also proved to identify disease-related genes (Chris Ding, 2005).

In the postgenomic era, one of the most important challenges is to determine gene functions. We use high-throughput biotechnologies to generate a large amount of biological data that are valuable resources for acquiring the gene functions (Li et al., 2012). For example, in order to predict protein functions, Nabieva et al. (2005) developed a network-flow-based algorithm that exploits the underlying structure of protein interaction maps. Now, protein–protein interaction (PPI) data have been widely used for gene function prediction.

The Search Tool for the Retrieval of Interacting Genes (STRING) database is an online resource that provides both experimental and predicted interaction information with a confidence score (Szklarczyk et al., 2011). In the PPI network, proteins with short distances between each other tend to have similar biological functions (Sharan et al., 2007; Bogdanov and Singh, 2010; Kourmpetis et al., 2010; Ng et al., 2010), and interactive neighbors are prone to have more similar biological functions than noninteractive ones (Letovsky and Kasif, 2003; Karaoz et al., 2004).

This study identified HCC-related genes based on the gene expression profile. We identified 117 HCC-related gene probes with the mRMR approach. We also used Dijkstra's algorithm (Dijkstra, 1959) to identify 187 genes on the shortest paths between the 117 gene probes in the PPI network. Then, the genes identified were analyzed through gene ontology (GO) functional enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment.

2. Materials and Methods

2.1. Dataset

The gene expression profiling dataset contains a total of 43 tumor and 52 nontumor samples, which were from research by Tsuchiya et al. (2010). It was generated by the Agilent-014850 Whole Human Genome Microarray 4x44K G4112F. The dataset was downloaded from NCBI Gene Expression Omnibus (GEO) with the accession number GSE17856.

2.2. Feature selection

We applied the mRMR method to rank the importance of the features, which is based on both their relevance to the target and the redundancy between features (Chris Ding, 2005). In our study, mRMR was realized by using an R package mRMRe (De Jay et al., 2013). In mRMRe, both relevance and redundancy are quantified using mutual information (MI), which is estimated as \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}I (x , y) = - \frac {1} {2} \ln [ 1 - \rho (x , y) ^2 ] \tag {1} \end{align*} \end{document}

where x and y are two features, I and ρ represent the MI and the correlation of x and y, respectively.

Let \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $$X = \{x_1 , \ldots , x_n \} $$ \end{document} denote the set of input features (gene probes), and let y denote the input target (phenotype). Given x_i as the feature with the highest MI with the target, then the set of ranked features, denoted by S, is initialized with x_i. Next, the remaining feature x_j with the best balance between maximal relevance and minimum redundancy is added to S. It is selected by maximizing the score q according to the following equation: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}q = I (x_j , y) - \frac {1} {\mid s \mid} \sum\nolimits_ {x_ {k} \in S} I (x_j , x_k) \tag {2} \end{align*} \end{document}

The selection step is repeated until a desired number of ranked features N has been reached.

The mRMR provides only a list of ranked features, but does not tell us how many appropriate features should be selected. So a possible feature subset S_i can be determined by incremental feature selection (IFS), as expressed by the following equation (Huang et al., 2009): \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}S_i = \{f_1 , f_2 , \ldots , f_i \} (1 \le i \le {\rm N}) \tag{3}\end{align*} \end{document}

For example, the first feature subset is S₁ = {f₁}, and the last feature subset is \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $$S_N = \{f_1 , f_2 , \ldots , f_N \} $$ \end{document} , which includes all the features.

2.3. Prediction engine

We used three prediction methods to estimate the phenotype of an individual: (1) the phenotype of its nearest neighbor, (2) the most frequent phenotypes of its five nearest neighbors, and (3) the phenotype of its nearest cluster center of every phenotypic group.

In our study, the distance between two individuals is defined according to Chou and coworkers (Chou and Shen, 2006; Chou, 2011), as shown below: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}D (i_1 , i_2) = 1 - \frac {e_1 \cdot e_2} {\mid e_1 \mid \cdot \mid e_2 \mid} \tag {4} \end{align*} \end{document}

where i₁ and i₂ represent two individuals, d refers to the distance between the two individuals, and e₁ and e₂ are vectors of selected feature sets (expression levels of selected genes) of the two individuals, respectively.

2.4. Validation method

Independent dataset test, subsampling test, and jackknife test are three validation methods that are often used in statistics for validating a statistical prediction. However, among the three validation methods, the jackknife test can avoid the arbitrariness problem that is existent in the independent dataset and subsampling test (Chou, 1995, 2011; Chou and Shen, 2008). In the jackknife test, both the training dataset and testing dataset are open, and each sample will be in turn moved between the training dataset and testing dataset.

The prediction accuracy was formulated by \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*} {\rm Accuracy} = \frac{{\rm TP} + {\rm TN}} {{\rm TP} + {\rm TN} + {\rm FP} + {\rm FN}} \tag {5} \end{align*} \end{document}

where TP represents the number of true positive, TN represents the number of true negative, FP represents the number of false positive, and FN represents the number of false negative.

2.5. Graphics approach and shortest paths tracing

In this study, we constructed a graph G(V,E) using the PPI data from STRING (version 9.0) (Szklarczyk et al., 2011), which can provide intuitive insights and overall structure properties to study complex biological systems. Based on our PPI network, we used Dijkstra's algorithm to identify the shortest path between any pair of proteins that were identified by mRMR-IFS, and we used Cytoscape (Smoot et al., 2011) to visualize the subnetwork of shortest paths.

2.6. Pathway enrichment analysis

In our study, we used the functional annotation tool of DAVID (Huang da et al., 2009a) for KEGG pathway enrichment and GO functional enrichment analysis. The original enrichment p-value and corrected p-value by Benjamin multiple tests were estimated by the tool.

3. Results

3.1. Gene probes for HCC classification identified by mRMR

We retrieved the HCC gene expression profile from GEO with the accession number GSE17856, which contained 95 samples (43 tumors, 52 nontumors) and 25,073 probes. We applied the mRMR-IFS strategy for feature selection and compared three different nearest-neighbor models for phenotype classification (see Materials and Methods). On the basis of outputs of mRMR, we estimated the classification accuracy for these models for up to 500 probes using jackknife validation (Fig. 1). From the IFS curve, we find that although the feature number with the highest accuracy (0.905) is 405, the second accurate one is 117 and the accuracy (0.895) is almost similar to the highest one. Since fewer features were incorporated, we chose the 117 gene probes as the final feature set to separate tumor and nontumor samples. All the 117 gene probes and their corresponding genes are presented in Supplementary Table S1 (Supplementary Material is available online at www.liebertpub.com/cmb). Some of the genes have been shown to be closely related to HCC, such as MT1X, BMI1, and CAP2, which justified the rationality of our method. We also identified some cancer genes that have not been reported previously to be closely related to the HCC, such as TACSTD2, suggesting that our method is effective in prediction of novel HCC-related genes. And their expression difference is shown in Figure 2.

FIG. 1.

IFS curve to determine the number of features used in prediction. We used an IFS curve to determine the number of features finally used in the mRMR selection. Prediction accuracy reached its second maximum value when 117 genes were included. Predict 1, predict 2, and predict 3 refer to the three prediction methods we used—a vote of the first nearest neighbor, the top five nearest neighbor, and nearest clustering center of each phenotype group, respectively. The x-axis indicates the number of probes used for classification, and the y-axis is the prediction accuracy of the three approaches to predict samples' phenotype. IFS, incremental feature selection; mRMR, maximum relevance–minimum redundancy.

FIG. 2.

Expression differences of MT1X (A_23_P303242, A_23_P303242), BMI1 (A_23_P314115,A_23_P115732,A_24_P303989), CAP2 (A_23_P421664), and TACSTD2 (A_23_P149529) between tumors and nontumors. This figure shows the expression differences of MT1X, BMI1, CAP2, and TACSTD2 between tumors and nontumors separately. Error bars indicate standard errors.

3.2. A PPI subnetwork of the identified genes on HCC

We further constructed an undirected graph using the PPI data from STRING (Szklarczyk et al., 2011). From the 117 gene probes identified by mRMR and IFS as described above, we found 75 protein coding genes, and 32 proteins of them existed in STRING. Then, we picked out every pair of two proteins from the 32 proteins and identified the shortest path between these two proteins with Dijkstra's algorithm (Dijkstra, 1959). The shortest paths between the 32 proteins were integrated into a subnetwork (Fig. 3), which contains a total of 350 PPIs involving 187 proteins. To evaluate the essentiality of the 187 proteins, we ranked them according to their betweennesses, and the top 10 proteins are shown in Table 1. Among these genes, the ubiquitin C (UBC) gene has the largest betweenness of 290, meaning that there are at least 145 shortest paths going through this gene. This result suggests that UBC may play important roles during the pathogenesis of HCC.

FIG. 3.

PPI network of shortest paths among 75 computational method-identified proteins. Shortest paths between each pair of the 32 proteins that were from the 75 computational method-selected proteins were identified in the STRING PPI network. Proteins are presented using their Ensembl IDs. Proteins in red are the 32 identified using the computational method that also present in the STRING PPI network; proteins in dark green are located only on shortest paths. PPI, protein–protein interaction; STRING, Search Tool for the Retrieval of Interacting Genes.

Table 1.

Top 10 of the Genes by Betweenness in the Shortest Paths

Ensembl gene ID	Associated gene symbol	Ensembl transcript ID	Ensembl protein ID	Betweenness
ENSG00000150991	UBC	ENST00000339647	ENSP00000344818	290
ENSG00000141510	TP53	ENST00000269305	ENSP00000269305	136
ENSG00000110395	CBL	ENST00000264033	ENSP00000264033	88
ENSG00000100644	HIF1A	ENST00000337138	ENSP00000338018	75
ENSG00000135679	MDM2	ENST00000462284	ENSP00000417281	74
ENSG00000163655	GMPS	ENST00000496455	ENSP00000419851	54
ENSG00000187555	USP7	ENST00000344836	ENSP00000343535	54
ENSG00000198931	APRT	ENST00000378364	ENSP00000367615	54
ENSG00000146648	EGFR	ENST00000275493	ENSP00000275493	51
ENSG00000115414	FN1	ENST00000354785	ENSP00000346839	37

3.3. The result of enrichment analysis of the identified genes

Using the functional annotation tool provided by DAVID (Huang da et al., 2009a,b), the KEGG pathway and GO functional enrichment analysis were first performed on the genes corresponding to the 117 gene probes. This result shows that these genes are significantly enriched in the GO modules of protein localization in organelles (Supplementary Table S2). In the KEGG pathway enrichment analysis, a pathway (hsa04080: neuroactive ligand–receptor interaction) is enriched, but it is not significant (p = 0.068).

The KEGG pathway and GO functional enrichment were also carried out for all the 187 genes on the shortest paths. In the GO functional enrichment analysis, many genes are significantly enriched in the modules related to cell death (Supplementary Table S3). The KEGG pathway enrichment result shows that these genes are significantly enriched in the pathways in cancer (p = 1.02E-26; Supplementary Table S4).

4. Discussion

In the 117 gene probes optimizing the tumor and nontumor prediction, both known and unknown HCC-related cancer genes are found. For example, a series of MT (metallothioneins) genes are included, such as MT1A, MT1B, MT1G, MT1H, MT1L, and MT1X. MTs are cysteine-rich low-molecular-mass intracellular proteins occurring in a wide variety of eukaryotes and constituting the major fraction of intracellular protein thilos (Jaromir Gumulec et al., 2014). Many studies have shown an increased expression of MTs in human tumors of breast (Schmid et al., 1993; Goulding et al., 1995; Jin et al., 2001); colon (Dziegiel, 2003); kidney (Nartey et al., 1987a); liver (Nartey et al., 1987a); lung (Joseph et al., 2001); nasopharynx (Jayasurya et al., 2000); ovary (Tan et al., 1999); prostate (Garrett et al., 2000); salivary gland, testes, and thyroid (Nartey et al., 1987b); and urinary bladder (Cherian, 2003). In 1997, Studer et al. also found that metallothionein accretion in human hepatic cells is linked to cellular proliferation (Studer et al., 1997). MT1X is not only found by mRMR-IFS, but also appears in the genes that exist in the subnetwork consisting of shortest paths. In a previous study, a cDNA microarray dataset on HCC indicated that MT1X shows downregulated expression (Peng et al., 2012). Another study also found that the downregulated expression of MT in HCC may play a role in hepatocarcinogenesis and be a marker of hepatocellular differentiation (Tao et al., 2007). Our study confirms that MT1X is downexpressed and is very relevant to the pathogenesis of HCC.

Another gene among the 117 probes we noticed is B lymphoma Mo-MLV insertion region 1 homolog (BMI1), which is known as a polycomb ring finger oncogene. Chatoo et al. (2009) found that the polycomb group gene BMI1 regulates antioxidant defenses in neurons by repressing p53 pro-oxidant activity. Wang et al. (2008a) also found that increased polycomb-group oncogene BMI1 expression correlated with poor prognosis in HCC. Our study also supported this opinion. In fact, BMI1 is not only well established as one of the most commonly activated oncogenes, but also provides a potential therapeutic target for the future treatment of HCC.

Adenylyl cyclase-associated protein 2 is an enzyme that in humans is encoded by the CAP2 gene, which was identified by mRMR-IFS in our study. Neo et al. (2004) identified several genes that were upregulated in human HCCs using gene expression profiling, and CAP2 was listed as one of the overexpressed genes. In Shibata et al.'s study, they also found that CAP2 was overexpressed at both the mRNA and protein levels in early and progressed HCCs (Shibata et al., 2006). In our study, we also found that the CAP2 gene is overexpressed in HCC, which provided a new evidence that CAP2 is related to the pathogenesis of HCC.

Tumor-associated calcium signal transducer 2 (TACSTD2; also called Trop-2) is a calcium signal transducer (Ripani et al., 1998) that is associated with cancer aggressiveness, poor prognosis (Fong et al., 2008a; Muhlmann et al., 2009), and transformed cell growth in experimental systems (Wang et al., 2008b). Trerotola et al. (2013) found that upregulation of trop-2 quantitatively stimulated human cancer growth. There are many cancers that are associated with the expression of Trop-2, such as pancreatic, gastric, oral, colorectal, and ovarian cancers (Fong et al., 2008a,b; Wang et al., 2008b; Muhlmann et al., 2009; Bignotti et al., 2010). The finding in our work that Trop-2 was identified in HCC prediction suggested its potential role in the development of HCC.

In the PPI network constructed by shortest paths tracing, more HCC-related genes could be revealed; especially UBC has the highest betweenness in the subnetwork. We did not find any previous reports about the effect of UBC on the development of HCC, but ubiquitins are related to the DNA-damage responses as well as signaling processes leading to activation of the transcription factor NF-kappa-B (Chen, 2005). Thus, our results imply that UBC may play important roles during the pathogenesis of HCC.

5. Conclusions

We applied a set of machine learning approaches, including mRMR and IFS, to the HCC gene expression profile and identified 117 gene probes for HCC prediction. In these genes, some have been shown to be closely related to HCC (e.g., MT1X, BMI1, and CAP2). We also identified some new genes that have not been reported previously to be closely related to the process of HCC (e.g., TACSTD2), which bring some new insights into the occurrence of HCC. We further constructed a PPI subnetwork based on the selected genes using shortest paths tracing and identified 187 genes relevant to HCC. Functional enrichment analysis shows that the subnetwork is significantly enriched in modules related to cell death and pathways in cancer. Network analysis reveals the potential role of UBC in the pathogenesis of HCC. We demonstrated that our method is efficient and effective in the prediction of novel cancer-related genes and has universal applicability in cancer research.

Authors' Contributions

T.G., X.D., R.L., Z.W., and Y.L. wrote the article. T.G., X.D., and Z.W. performed the analysis. Y.L. organized the whole project.

Footnotes

Acknowledgments

This work was supported by the National Natural Science Foundation of China (31260544, U1036604, 31070752, and 31301032).

Author Disclosure Statement

No competing financial interests exist.

References

Bignotti

, Todeschini

, Calza

, et al. 2010. Trop-2 overexpression as an independent marker for poor overall survival in ovarian carcinoma patients. Eur. J. Cancer, 46, 944–953.

Bogdanov

, and Singh

A.K.

2010. Molecular function prediction using neighborhood features. IEEE/ACM Trans. Comput. Biol. Bioinform., 7, 208–217.

Chatoo

, Abdouh

, David

, et al. 2009. The polycomb group gene Bmi1 regulates antioxidant defenses in neurons by repressing p53 pro-oxidant activity. J. Neurosci., 29, 529–542.

Chen

Z.J.

2005. Ubiquitin signalling in the NF-κB pathway. Nat. Cell Biol., 7, 758–765.

Cherian

2003. Metallothioneins in human tumors and potential roles in carcinogenesis. Mut. Res., 533, 201–209.

Chou

K.-C.

1995. Prediction of protein structural classes. Biochem. Mol. Biol., 30, 275–349.

Chou

K.C.

2011. Some remarks on protein attribute prediction and pseudo amino acid composition. J. Theor. Biol., 273, 236–247.

Chou

K.C.

, and Shen

H.B.

2006. Predicting eukaryotic protein subcellular location by fusing optimized evidence-theoretic k-nearest neighbor classifiers. J. Proteome Res., 5, 1888–1897.

Chou

K.C.

, and Shen

H.B.

2008. Cell-PLoc: a package of Web servers for predicting subcellular localization of proteins in various organisms. Nat. Protoc., 3, 153–162.

10.

Chris Ding

H.P.

2005. Minimum redundancy feature selection from microarray gene expression data. J. Bioinform. Comput. Biol., 3, 185–205.

11.

De Jay

, Papillon-Cavanagh

, Olsen

, et al. 2013. mRMRe: an R package for parallelized mRMR ensemble feature selection. Bioinformatics, 29, 2365–2368.

12.

Dijkstra

E.W.

1959. A note on two problems in connexion with graphs. Num. Math., 1, 269–271.

13.

Dziegiel

, Forgacz

, and Suder

, et al. 2003. Prognostic significance of metallothionein expression in correlation with Ki-67 expression in adenocarcinoma of large intestine. Histol. Histopathol., 18, 401–407.

14.

El-Serag

H.B.

, and Rudolph

K.L.

2007. Hepatocellular carcinoma: epidemiology and molecular carcinogenesis. Gastroenterology, 132, 2557–2576.

15.

Fong

, Moser

, Krammel

, et al. 2008a. High expression of TROP2 correlates with poor prognosis in pancreatic cancer. Br. J. Cancer, 99, 1290–1295.

16.

Fong

, Spizzo

, Gostner

J.M.

, et al. 2008b. TROP2: a novel prognostic marker in squamous cell carcinoma of the oral cavity. Mod. Pathol., 21, 186–191.

17.

Garrett

S.H.

, Sens

M.A.

, and Shukla

, et al. 2000. Metallothionein isoform 1 and 2 gene expression in the human prostate: downregulation of MT-1X in advanced prostate cancer. Prostate, 43, 125–135.

18.

Goulding

, Jasani

, Pereira

, et al. 1995. Metallothionein expression in human breast cancer. Br. J. Cancer, 72, 968–972.

19.

Huang

, Cui

, Hu

, et al. 2009. Prediction of pharmacological and xenobiotic responses to drugs based on time course gene expression profiles. PloS ONE, 4, e8126.

20.

Huang da

, Sherman

B.T.

, and Lempicki

R.A.

2009a. Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res., 37, 1–13.

21.

Huang da

, Sherman

B.T.

, and Lempicki

R.A.

2009b. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc., 4, 44–57.

22.

Jaromir Gumulec

M.R.

, Vojtech

, Rene

, and Michal

2014. Metallothionein–immunohistochemical cancer biomarker: a meta-analysis. PloS ONE, 9, e85346.

23.

Jayasurya

, Bay

B.H.

, Yap

W.M.

, et al. 2000. Proliferative potential in nasopharyngeal carcinoma: correlation with metallothionein expression and tissue zinc levels. Carcinogenesis, 21, 1809–1812.

24.

Jemal

, Bray

, Center

M.M.

, et al. 2011. Global cancer statistics. CA Cancer J. Clin., 61, 69–90.

25.

Jin

, Bay

B.H.

, Chow

V.T.

, and Tan

P.H.

2001. Metallothionein 1F mRNA expression correlates with histological grade in breast carcinoma. Breast Cancer Res. Treat., 66, 265–272.

26.

Joseph

M.G.

, Banerjee

, Kocha

, et al. 2001. Metallothionein expression in patients with small cell carcinoma of the lung: correlation with other molecular markers and clinical outcome. Cancer, 92, 836–842.

27.

Karaoz

, Murali

T.M.

, Letovsky

, et al. 2004. Whole-genome annotation by using evidence integration in functional-linkage networks. Proc. Natl. Acad. Sci. USA, 101, 2888–2893.

28.

Kourmpetis

Y.A.I.

, van Dijk

A.D.J.

, Bink

M.C.A.M.

, et al. 2010. Bayesian Markov random field analysis for protein function prediction based on network data. PloS ONE, 5, e9293.

29.

Letovsky

, and Kasif

2003. Predicting protein function from protein/protein interaction data: a probabilistic approach. Bioinformatics, 19, i197–i204.

30.

B.Q.

, Zhang

, Huang

, et al. 2012. Identification of retinoblastoma related genes with shortest path in a protein-protein interaction network. Biochimie, 94, 1910–1917.

31.

Llovet

J.M.

, Burroughs

, and Bruix

2003. Hepatocellular carcinoma. Lancet, 362, 1907–1917.

32.

Muhlmann

, Spizzo

, Gostner

, et al. 2009. TROP2 expression as prognostic marker for gastric carcinoma. J. Clin. Pathol., 62, 152–158.

33.

Nabieva

, Jim

, Agarwal

, et al. 2005. Whole-proteome prediction of protein function via graph-theoretic analysis of interaction maps. Bioinformatics, 21 Suppl 1, i302–i310.

34.

Nartey

N.O.

, Banerjee

, and Cherian

M.G.

1987a. Immunohistochemical localization of metallothionein in cell nucleus and cytoplasm of fetal human liver and kidney and its changes during development. Pathology, 19, 233–238.

35.

Nartey

, Cherian

M.G.

, and Banerjee

1987b. Immunohistochemical localization of metallothionein in human thyroid tumors. Am. J. Pathol., 129, 177–182.

36.

Neo

S.Y.

, Leow

C.K.

, Vega

V.B.

, et al. 2004. Identification of discriminators of hepatoma by gene expression profiling using a minimal dataset approach. Hepatology, 39, 944–953.

37.

K.L.

, Ciou

J.S.

, and Huang

C.H.

2010. Prediction of protein functions based on function-function correlation relations. Comput. Biol. Med., 40, 300–305.

38.

Peng

, Gu

, Xiong

, et al. 2012. Microarray-assisted pathway analysis identifies MT1X & NFkappaB as mediators of TCRP1-associated resistance to cisplatin in oral squamous cell carcinoma. PloS ONE, 7, e51413.

39.

Ripani

, Sacchetti

, Corda

, and Alberti

1998. Human TROP-2 is a tumor-associated calcium signal transducer. Int. J. Cancer, 76, 671–676.

40.

Schmid

K.W.

, Ellis

, Gee

J.M.

, et al. 1993. Presence and possible significance of immunocytochemically demonstrable metallothionein over-expression in primary invasive ductal carcinoma of the breast. Virchows Arch. A Pathol. Anat. Histopathol., 422, 153–159.

41.

Sharan

, Ulitsky

, and Shamir

2007. Network-based prediction of protein function. Mol. Syst. Biol., 3, 88.

42.

Shibata

, Mori

, Du

, et al. 2006. Overexpression of cyclase-associated protein 2 in multistage hepatocarcinogenesis. Clin. Cancer Res., 12, 5363–5368.

43.

Smoot

M.E.

, Ono

, Ruscheinski

, et al. 2011. Cytoscape 2.8: new features for data integration and network visualization. Bioinformatics, 27, 431–432.

44.

Studer

, Vogt

C.P.

, Cavigelli

, et al. 1997. Metallothionein accretion in human hepatic cells is linked to cellular proliferation. Biochem. J., 328, 63–67.

45.

Szklarczyk

, Franceschini

, Kuhn

, et al. 2011. The STRING database in 2011: functional interaction networks of proteins, globally integrated and scored. Nucleic Acids Res. 39, D561–D568.

46.

Tan

, Sinniah

, Bay

B.H.

, and Singh

1999. Metallothionein expression and nuclear size in benign, borderline, and malignant serous ovarian tumours. J. Pathol., 189, 60–65.

47.

Tao

, Zheng

J.M.

, Xu

A.M.

, et al. 2007. Downregulated expression of metallothionein and its clinicopathological significance in hepatocellular carcinoma. Hepatol. Res., 37, 820–827.

48.

Trerotola

, Cantanelli

, Guerra

, et al. 2013. Upregulation of Trop-2 quantitatively stimulates human cancer growth. Oncogene, 32, 222–233.

49.

Tsuchiya

, Parker

J.S.

, Kono

, et al. 2010. Gene expression in nontumoral liver tissue and recurrence-free survival in hepatitis C virus-positive hepatocellular carcinoma. Mol. Cancer, 9, 74.

50.

Wang

, Pan

, Zhang

H.K.

, et al. 2008a. Increased polycomb-group oncogene Bmi-1 expression correlates with poor prognosis in hepatocellular carcinoma. J. Cancer Res. Clin. Oncol., 134, 535–541.

51.

Wang

, Day

, Dong

, et al. 2008b. Identification of Trop-2 as an oncogene and an attractive therapeutic target in colon cancers. Mol. Cancer Ther., 7, 280–285.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.01 MB

0.35 MB

0.02 MB