Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

HPOSim: An R Package for Phenotypic Similarity Measure and Enrichment Analysis Based on the Human Phenotype Ontology

  • Yue Deng,

    Affiliations School of Computer Science and Technology, Xidian University, Xi'an, People’s Republic of China, Institute of Software Engineering, Xidian University, Xi'an, People’s Republic of China

  • Lin Gao ,

    Affiliation School of Computer Science and Technology, Xidian University, Xi'an, People’s Republic of China

  • Bingbo Wang,

    Affiliation School of Computer Science and Technology, Xidian University, Xi'an, People’s Republic of China

  • Xingli Guo

    Affiliation School of Computer Science and Technology, Xidian University, Xi'an, People’s Republic of China

HPOSim: An R Package for Phenotypic Similarity Measure and Enrichment Analysis Based on the Human Phenotype Ontology

  • Yue Deng, 
  • Lin Gao, 
  • Bingbo Wang, 
  • Xingli Guo



Phenotypic features associated with genes and diseases play an important role in disease-related studies and most of the available methods focus solely on the Online Mendelian Inheritance in Man (OMIM) database without considering the controlled vocabulary. The Human Phenotype Ontology (HPO) provides a standardized and controlled vocabulary covering phenotypic abnormalities in human diseases, and becomes a comprehensive resource for computational analysis of human disease phenotypes. Most of the existing HPO-based software tools cannot be used offline and provide only few similarity measures. Therefore, there is a critical need for developing a comprehensive and offline software for phenotypic features similarity based on HPO.


HPOSim is an R package for analyzing phenotypic similarity for genes and diseases based on HPO data. Seven commonly used semantic similarity measures are implemented in HPOSim. Enrichment analysis of gene sets and disease sets are also implemented, including hypergeometric enrichment analysis and network ontology analysis (NOA).


HPOSim can be used to predict disease genes and explore disease-related function of gene modules. HPOSim is open source and freely available at SourceForge (


Phenotypic similarity plays an important role in different biological and biomedical applications. Previous studies prove that genes with similar phenotypes yields biological modules in terms of diseases, thus it can be used in predicting disease-causing genes [1][2]. Furthermore, it is crucial for understanding the relationships between different diseases [3].

Most current methods for measuring phenotypic similarity [4][5] are based on the Online Mendelian Inheritance in Man (OMIM) database [6] that contains textual records representing genetic disorders. However, the absence of a controlled vocabulary makes it difficult to analyze the OMIM data using a computational approach [7]. The Human phenotype ontology (HPO) [8] provides a controlled and standardized vocabulary of phenotypic abnormalities annotating all clinical entries in OMIM, which sheds light on the large-scale computational analysis of the human phenome, i.e., DECIPHER [9], ECARUCA [10] and Bridge [11].

Several tools using HPO-based semantic similarity are currently available. Phenomizer [12] is the first tool for semantic similarity search over HPO, in which users input the phenotypic abnormalities of a patient as HPO IDs, and obtain a list of diagnoses in OMIM IDs. Other tools include OwlSim [13], PhenoDigm [14], PhenomeNET/PhenomeBrowser [15] and OntoSIML [16]. The detailed comparison of HPOSim and other HPO-based tools is shown in Table 1. It can be seen from the table that most of the existing tools share one drawback: the calculations of phenotypic similarity for HPO terms, genes and diseases are not well supported. Although OntoSIML and OwlSim provide functions for calculating semantic similarity, users are required to manually input the mapping from entities (gene or disease) to HPO terms, which entails additional preprocessing effort.

In addition, there exist several tools for HPO-based enrichment analysis. OntoFUNC [17] performs functional enrichment analysis over ontologies in OWL format. It is based on FUNC [18] and users need to manually input the mapping data, which is the same as OntoSIML. STOP [19] is an online tool and can be used as a Cytoscape plug-in. It can be used in the enrichment analysis of gene sets, but does not support the analysis of disease set.

Several R packages for semantic similarity and enrichment analysis are available, including GOSim [20], GOSemSim [21], DOSim [22], DOSE [23] and topGO [24]. However, these packages mainly use gene ontology (GO) [25] and disease ontology (DO) [26]. To the best of our knowledge, there is no R package that focuses on HPO-based semantic similarity and enrichment analysis.

Thus, we developed an R package HPOSim with an immediate purpose to capturing phenotypic similarities between genes and diseases. The framework of HPOSim is shown in Fig. 1. HPOSim analyzes semantic similarity for HPO terms, genes and diseases. Functional enrichment analysis of gene set and disease set are also provided, including the classic hypergeometric enrichment analysis and the novel network ontology analysis (NOA) [27].

Figure 1. Framework of HPOSim.

Users can use HPOSim to calculate semantic similarity for HPO terms, genes and diseases. HPOSim can also be used to identify enriched HPO terms for gene set and disease set.



HPO contains over 10000 terms (10686 terms in the HPO build #1042 released in September 2014) in three sub-ontologies, which are phenotypic abnormality (PA), onset and clinical course (OC) and mode of inheritance (MI). Approximately 99% of the HPO terms are in the PA sub-ontology. In each sub-ontology, terms are arranged in a directed acyclic graph (DAG) and are related to their parent terms by “is a” relationships. The structure of the HPO allows a term to have multiple parent terms, which enables different aspects of phenotypic abnormalities to be explored. Diseases and genes are annotated to the most specific terms possible, which means that if a disease or a gene is annotated to a term then all of the ancestors of this term also apply (see Fig. 2 for an example).

Figure 2. Example of the structure of HPO.

HPO term Abnormality of the joints of the lower limbs (HP:0100491) and all its ancestor terms are shown. Each term in the HPO describes a phenotypic abnormality. Terms are related to parent terms by “is a” relationships in the form of a directed acyclic graph. If a disease or a gene is annotated to a term, it will also be annotated to all of its ancestors.

The official ontology file provided by the HPO Consortium is in obo format, which is plain text-based. Thus, like other widely used R package for biomedical ontologies, e.g. GO.db, we constructed an R package termed HPO.db. HPO.db provided programmatic interfaces to the hierarchical structure of HPO terms. HPOSim uses HPO.db to obtain information about terms and relationships between terms. HPO.db can be used by other R packages that use HPO data.

HPOSim provides two kinds of pre-calculated data within the package: the association between HPO terms, as well as association between genes and diseases (gene-to-phenotype, phenotype-to-gene, disease-to-phenotype and phenotype-to-disease). The associations between HPO terms are obtained from the original ontology and annotation data provided by the HPO Consortium, and the information content (IC) of the HPO terms is pre-calculated based on both genes and diseases annotated to a certain term, while semantic similarity between genes and diseases are based on the IC of HPO terms.

The IC of a term t in HPO can be defined as follows:


where p(t) is the probability of observing t and its descendants in all genes/diseases annotated to a certain sub-ontology of HPO.

Measuring the similarity between HPO terms

Recently, several metrics that measure the semantic similarity between ontology annotations have been proposed [28]. In HPOSim, we implement seven commonly used semantic similarity measures to measure the similarity between HPO terms: the Resnik measure [29], Lin measure [30], Jiang–Conrath measure [31], relevance measure [32], information coefficient measure [33], graph IC measure [34] and Wang measure [35]. The first six measures are based on IC, while the Wang measure uses both IC and graph structure.

The Resnik measure defines the similarity between terms as the IC of their most informative common ancestor (MICA):


where tMICA is the MICA of term t1 and t2.

The Lin and Jiang–Conrath measures consider the IC of the two terms besides the IC of their MICA:


The relevance measure and the information coefficient measure are based on Lin’s measure:


The graph IC measure takes all the common ancestors of the two terms into account:


where A(t) is the ancestors of term t in HPO.

The Wang measure is based on the graph structure of HPO DAG. In Wang’s measure, a weight is given to each edge according to its type. DAGt = (t,Tt,Et) represents the subgraph made up of term t and its ancestors, where Tt is the set of the ancestor terms of t and Et is the set of edges in DAGt.

In DAGt, St(n) measures the semantic contribution of term n to term t, which is defined as:


The similarity between HPO term t1 and termt2 is defined as:


where SV(m) is the sum of the semantic contributions of all the terms in DAGm.

Combining term-term similarity into gene-gene and disease-disease similarity

In HPOSim, the similarity between two genes is calculated based on the pairwise similarity of the two HPO term sets annotating these two genes. HPOSim provides five methods to combine multiple term-term similarities into one gene-gene similarity, which are “Max” [36], “Mean” [36], “funSimMax” [32], “funSimAvg” [32], and “BMA” [35].

Given gene g1 annotated by HPO term set HPO1 = {t11, t12,…,t1m} and g2 annotated by HPO2 = {t21, t22,…,t2n}. The similarity matrix S=[si j]m×n contains all pairwise similarity scores of terms in HPO1 and HPO2.

The “Max” method calculates the maximum semantic similarity score over all pairs of HPO terms in the two term sets, and is defined as follows.


The “Mean” method calculates the average semantic similarity score over all pairs of HPO terms in the two term sets, and is defined as follows.


The “funSimMax”, “funSimAvg” and “BMA” methods are based on the maximum value in each row and column of similarity matrix S.

The “funSimMax” and “funSimAvg” methods [32] use the arithmetic maxima and average between similarities for two directional comparisons of the similarity matrix S.


The “BMA” method uses the best-match average strategy, which calculates the average of all maximum similarities on each row and column of the similarity matrix S.


The calculation of the similarity between diseases is the same as that between genes. The similarity between two diseases is calculated based on the pairwise similarity of the two term sets annotating these two diseases.

HPO-based Enrichment Analysis

HPOSim provides HPO-based enrichment analysis to investigate the phenotypic features of gene sets or disease sets. Two enrichment analysis methods are provided: hypergeometric test and the NOA method [27].

Given an HPO term t and a gene set with T genes, assuming that there are R genes/diseases annotated in the whole HPO in which G genes/diseases are annotated to t. In addition, there are O genes/diseases in the gene set that are annotated to t. The hypergeometric enrichment p-value for t is calculated as follows:


In NOA, users input a gene or disease network. For each edge in the network, the HPO terms annotating this edge are defined as the intersection of the two term sets annotating the two nodes of the edge. NOA uses HPO terms annotating the edges to perform the enrichment analysis. Two alternative strategies, “sub-net” and “whole-net”, are applied to choose the reference set. In the “sub-net” strategy, users are required to provide the reference set. While in the “whole-net” strategy, the complete graph on the nodes of the input network is used as the reference set.

Results and Discussion

HPOSim consists of two parts: (i) the similarity measures between phenotypes (HPO terms), between human genes (Entrez IDs) and between diseases (OMIM IDs), and (ii) HPO-based enrichment analysis (NOA and the hypergeometric method) for gene set and disease set.

Application on gene similarity and gene set enrichment analysis

We used the aging network [37] to demonstrate the application of gene semantic similarity provided by HPOSim. The aging network was constructed by identifying genes related to aging and adding edges between interacting gene pairs. After removing the genes that are not annotated in the PA sub-ontology of HPO, 102 genes and 293 interactions were remained (see S1 Dataset for detail).

First, the semantic similarity matrix of the 102 genes was constructed using the Resnik measure and “funSimMax” combining method (see S2 Dataset for detail). A hierarchical clustering was then performed using the R package stats, and six modules were detected using the R package dynamicTreeCut. HPO enrichment analysis (hypergeometric test) was then performed using HPOSim. GO enrichment analysis and pathway enrichment analysis based on KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway database [38] were performed using DAVID [39]. The results are shown in Table 2.

It can be seen that the enriched GO and HPO annotations are largely different among these modules. For example, the enriched GO annotations of module M2 implied that aging is associated with radiation including ultraviolet (UV), which has been verified by previous study in skin aging [40]. While the enriched GO annotations of module M3 implied that aging is associated with hormone stimulus, and literature mining showed that older women require a greater parathyroid hormone stimulus than younger women [41]. The enriched HPO annotations of the module M3 implied that aging are associated with abnormality of the pituitary, which has been verified by Sano et al. [42]. Disease enrichment analysis based on OMIM was then performed on genes in M3 using DAVID [39] and showed that term “Pituitary hormone deficiency, combined” was representative (p-value = 8.2E-3).

The enriched pathways of different modules are closely related to cancer, however various among different modules. Jak-STAT signaling pathway was found to be representative in modules M3 and M4. In a previous study by Fulop et al. [43], it was found that the signalling of IL-2 receptors is altered in T cells and macrophages with aging, mainly in relation to the Jak-STAT pathway.

These results above indicate that HPO-based semantic similarity can provide a different aspect in disease-related studies other than GO.

NOA and hypergeometric gene set enrichment analysis were then performed on the aging network. The “whole-net” strategy [27] was used to choose the reference set in NOA. The top 10 enriched HPO terms in the two enrichment methods are shown in Table 3. It can be seen that both enrichment methods identify neoplasm-related HPO terms as the top hits. However, these two methods give different terms and different ranks of terms. When dealing with gene/disease sets from biological networks, users are suggested to use the NOA method. If the gene sets are not from network data, users can use either hypergeometric or NOA enrichment method.

Table 3. Top 10 enriched HPO terms by the NOA method and hypergeometric enrichment.

Application on disease similarity and disease set enrichment analysis

HPOSim can also be used to investigate the phenotypic relationships between diseases. First, 115 cancer related entries were obtained by searching the OMIM database [6] using “cancer” or “carcinoma” as the key word. After removing the diseases that are not annotated in the PA sub-ontology of HPO and all the genes, 55 disease entries were remained (see S3 Dataset for detail).

The semantic similarity matrix of the 55 disease entries was constructed using the Resnik measure and “funSimMax” combining method (see S4 Dataset for detail). A hierarchical clustering was then performed and four modules were detected using the same routine as used in the previous case study. HPO enrichment analysis (hypergeometric test) was also performed using HPOSim. The results are shown in Table 4.

The results showed that these four disease modules had different phenotypic features. For example, module M3 included several types of women-only cancer, including breast cancer (OMIM:114480), breast-ovarian cancer (OMIM:604370, OMIM:612555), ovarian cancer (OMIM:167000) and cervical cancer(OMIM:603956). And lung cancer (OMIM:211980) in M3 was the second most commonly diagnosed types of cancer among women in 2013[44].

The result above indicated that HPO-based semantic similarity had potential ability to play an important role in disease classification and other disease-related studies.


HPOSim is an open source R package that contains seven semantic similarity measures and two enrichment analysis based on HPO data. Also, it provides useful functions for disease-related research and can be integrated with other R packages. In future work, we will integrate more similarity measures and other functions, such as visualization of the HPO data.

Supporting Information

S1 Dataset. Aging network after removing the genes that are not annotated in PA sub-ontology of HPO.


S2 Dataset. Semantic similarity matrix of the 102 genes in the aging network.


S4 Dataset. Semantic similarity matrix of the 55 cancer entries.



We thank the Human Phenotype Ontology Consortium for the original ontology and annotation data, and also thank Dr. Xiaoke Ma for reviewing the manuscript.

Author Contributions

Conceived and designed the experiments: YD LG BW XG. Performed the experiments: YD LG BW. Analyzed the data: YD BW XG. Contributed reagents/materials/analysis tools: YD XG. Wrote the paper: YD LG BW XG.


  1. 1. Köhler S, Bauer S, Horn D, Robinson PN (2008) Walking the Interactome for Prioritization of Candidate Disease Genes. Am J Hum Genet 82: 949–958. pmid:18371930
  2. 2. Van Driel MA, Bruggeman J, Vriend G, Brunner HG, Leunissen JAM (2006) A text-mining analysis of the human phenome. Eur J Hum Genet 14: 535–542. pmid:16493445
  3. 3. Oti M, Brunner H (2007) The modular nature of genetic diseases. Clinical Genetics 71: 1–11. pmid:17204041
  4. 4. Lage K, Karlberg EO, Størling ZM, Olason PI, Pedersen AG, et al. (2007) A human phenome-interactome network of protein complexes implicated in genetic disorders. Nat Biotechnol 25: 309–316. pmid:17344885
  5. 5. Freudenberg J, Propping P (2002) A similarity-based method for genome-wide prediction of disease-relevant human genes. Bioinformatics 18: S110–S115. pmid:12385992
  6. 6. Hamosh A, Scott AF, Amberger JS, Bocchini CA, McKusick VA (2005) Online Mendelian Inheritance in Man (OMIM), a knowledgebase of human genes and genetic disorders. Nucl Acids Res 33: D514–D517. pmid:15608251
  7. 7. Robinson PN, Köhler S, Bauer S, Seelow D, Horn D, et al. (2008) The Human Phenotype Ontology: A Tool for Annotating and Analyzing Human Hereditary Disease. The American Journal of Human Genetics 83: 610–615.
  8. 8. Köhler S, Doelken SC, Mungall CJ, Bauer S, Firth HV, et al. (2013) The Human Phenotype Ontology project: linking molecular biology and disease through phenotype data. Nucl Acids Res: gkt1026.
  9. 9. Firth HV, Richards SM, Bevan AP, Clayton S, Corpas M, et al. (2009) DECIPHER: Database of Chromosomal Imbalance and Phenotype in Humans Using Ensembl Resources. The American Journal of Human Genetics 84: 524–533.
  10. 10. Vulto-van Silfhout AT, van Ravenswaaij CMA, Hehir-Kwa JY, Verwiel ETP, Dirks R, et al. (2013) An update on ECARUCA, the European Cytogeneticists Association Register of Unbalanced Chromosome Aberrations. Eur J Med Genet 56: 471–474. pmid:23851227
  11. 11. The BRIDGE Project. Available: Accessed 2014 December 10.
  12. 12. Köhler S, Schulz MH, Krawitz P, Bauer S, Dölken S, et al. (2009) Clinical Diagnostics in Human Genetics with Semantic Similarity Searches in Ontologies. The American Journal of Human Genetics 85: 457–464.
  13. 13. Washington NL, Haendel MA, Mungall CJ, Ashburner M, Westerfield M, et al. (2009) Linking Human Diseases to Animal Models Using Ontology-Based Phenotype Annotation. PLoS Biol 7. pmid:19956802
  14. 14. Smedley D, Oellrich A, Kohler S, Ruef B, Sanger Mouse Genetics Project, et al. (2013) PhenoDigm: analyzing curated annotations to associate animal models with human diseases. Database 2013: bat025–bat025. pmid:23660285
  15. 15. Hoehndorf R, Schofield PN, Gkoutos GV (2011) PhenomeNET: a whole-phenome approach to disease gene discovery. Nucl Acids Res 39: e119–e119. pmid:21737429
  16. 16. OntoSIML. Available: Accessed 2014 December 10.
  17. 17. Hoehndorf R, Dumontier M, Gkoutos GV (2012) Identifying aberrant pathways through integrated analysis of knowledge in pharmacogenomics. Bioinformatics 28: 2169–2175. pmid:22711793
  18. 18. Prüfer K, Muetzel B, Do H-H, Weiss G, Khaitovich P, et al. (2007) FUNC: a package for detecting significant associations between gene sets and ontological annotations. BMC Bioinformatics 8: 41. pmid:17284313
  19. 19. Wittkop T, TerAvest E, Evani US, Fleisch KM, Berman AE, et al. (2013) STOP using just GO: a multi-ontology hypothesis generation tool for high throughput experimentation. BMC Bioinformatics 14: 53. pmid:23409969
  20. 20. Frohlich H, Speer N, Poustka A, BeiSZbarth T (2007) GOSim—an R-package for computation of information theoretic GO similarities between terms and gene products. BMC Bioinformatics 8: 166. pmid:17519018
  21. 21. Yu G, Li F, Qin Y, Bo X, Wu Y, et al. (2010) GOSemSim: an R package for measuring semantic similarity among GO terms and gene products. Bioinformatics 26: 976–978. pmid:20179076
  22. 22. Li J, Gong B, Chen X, Liu T, Wu C, et al. (2011) DOSim: An R package for similarity between diseases based on Disease Ontology. BMC Bioinformatics 12: 266. pmid:21714896
  23. 23. DOSE: Disease Ontology Semantic and Enrichment analysis. Available: Accessed 2014 December 10.
  24. 24. Alexa A, Rahnenfuhrer J (2010) topGO: topGO: Enrichment analysis for Gene Ontology. topGO: topGO: Enrichment analysis for Gene Ontology. Available: Accessed 2014 December 10.
  25. 25. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, et al. (2000) Gene Ontology: tool for the unification of biology. Nat Genet 25: 25–29. pmid:10802651
  26. 26. Schriml LM, Arze C, Nadendla S, Chang Y-WW, Mazaitis M, et al. (2011) Disease Ontology: a backbone for disease semantic integration. Nucleic Acids Research 40: D940–D946. pmid:22080554
  27. 27. Wang J, Huang Q, Liu Z-P, Wang Y, Wu L-Y, et al. (2011) NOA: a novel Network Ontology Analysis method. Nucleic Acids Res 39: e87. pmid:21543451
  28. 28. Pesquita C, Faria D, Falcão AO, Lord P, Couto FM (2009) Semantic Similarity in Biomedical Ontologies. PLoS Comput Biol 5: e1000443. pmid:19649320
  29. 29. Resnik P (1995) Using information content to evaluate semantic similarity in a taxonomy. Proceedings of the 14th international joint conference on Artificial intelligence—Volume 1. IJCAI’95. San Francisco, CA, USA: Morgan Kaufmann Publishers Inc. pp. 448–453. Available: = 1625855.1625914. Accessed 2014 December 10.
  30. 30. Lin D (1998) An Information-Theoretic Definition of Similarity. Proceedings of the Fifteenth International Conference on Machine Learning. ICML’98. San Francisco, CA, USA: Morgan Kaufmann Publishers Inc. pp. 296–304. Available: = 645527.657297. Accessed 2014 December 10.
  31. 31. Jiang J, Conrath D (1997) Semantic Similarity Based on Corpus Statistics and Lexical Taxonomy. International Conference Research on Computational Linguistics (ROCLING X). p. 9008. Available: = 1997cmp.lg….9008J. Accessed 2014 December 10.
  32. 32. Schlicker A, Domingues F, Rahnenführer J, Lengauer T (2006) A new measure for functional similarity of gene products based on Gene Ontology. BMC Bioinformatics 7: 302. pmid:16776819
  33. 33. Li B, Wang JZ, Feltus FA, Zhou J, Luo F (2010) Effectively integrating information content and structural relationship to improve the GO-based similarity measure between proteins. arXiv:10010958. Available: Accessed 2014 December 10.
  34. 34. Pesquita C, Faria D, Bastos H, Falcão AO, Couto FM (2007) Evaluating GO-based Semantic Similarity Measures. Proceedings of the 10th Annual Bio-Ontologies Meeting. pp. 37–40. Available: Accessed 2014 December 10.
  35. 35. Wang JZ, Du Z, Payattakool R, Yu PS, Chen C-F (2007) A new method to measure the semantic similarity of GO terms. Bioinformatics 23: 1274–1281. pmid:17344234
  36. 36. Lord PW, Stevens RD, Brass A, Goble CA (2003) Investigating semantic similarity measures across the Gene Ontology: the relationship between sequence and annotation. Bioinformatics 19: 1275–1283. pmid:12835272
  37. 37. Wang J, Zhang S, Wang Y, Chen L, Zhang X-S (2009) Disease-Aging Network Reveals Significant Roles of Aging Genes in Connecting Genetic Diseases. PLoS Comput Biol 5: e1000521. pmid:19779549
  38. 38. Kanehisa M, Goto S (2000) KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Research 28: 27–30. pmid:10592173
  39. 39. Huang DW, Sherman BT, Lempicki RA (2008) Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protocols 4: 44–57.
  40. 40. Fisher GJ, Wang Z, Datta SC, Varani J, Kang S, et al. (1997) Pathophysiology of Premature Skin Aging Induced by Ultraviolet Light. New England Journal of Medicine 337: 1419–1429. pmid:9358139
  41. 41. Silverberg SJ, Shane E, de la Cruz L, Segre GV, Clemens TL, et al. (1989) Abnormalities in Parathyroid Hormone Secretion and 1,25-Dihydroxyvitamin D3 Formation in Women with Osteoporosis. New England Journal of Medicine 320: 277–281. pmid:2911322
  42. 42. Sano T, Kovacs KT, Scheithauer BW, Young WF (1993) Aging and the Human Pituitary Gland. Mayo Clinic Proceedings 68: 971–977. pmid:8412363
  43. 43. Fulop T, Larbi A, Douziech N, Levesque I, Varin A, et al. (2006) Cytokine receptor signalling and aging. Mechanisms of Ageing and Development 127: 526–537. pmid:16530252
  44. 44. Siegel R, Naishadham D, Jemal A (2013) Cancer statistics, 2013. CA: A Cancer Journal for Clinicians 63: 11–30.