A literature-driven method to calculate similarities among diseases

Highlights

• We proposed a novel method which calculates disease–disease similarity.

• The proposed method can be used in constructing of disease network.

• The proposed method discovered the largest number of answer disease pairs and showed lowest p-value when compared with other comparable methods.

• The proposed method can provide an insight of relationship between diseases.

Abstract

Background

“Our lives are connected by a thousand invisible threads and along these sympathetic fibers, our actions run as causes and return to us as results”. It is Herman Melville's famous quote describing connections among human lives. To paraphrase the Melville's quote, diseases are connected by many functional threads and along these sympathetic fibers, diseases run as causes and return as results. The Melville's quote explains the reason for researching disease–disease similarity and disease network. Measuring similarities between diseases and constructing disease network can play an important role in disease function research and in disease treatment. To estimate disease–disease similarities, we proposed a novel literature-based method.

Methods and results

The proposed method extracted disease–gene relations and disease–drug relations from literature and used the frequencies of occurrence of the relations as features to calculate similarities among diseases. We also constructed disease network with top-ranking disease pairs from our method. The proposed method discovered a larger number of answer disease pairs than other comparable methods and showed the lowest p-value.

Conclusions

We presume that our method showed good results because of using literature data, using all possible gene symbols and drug names for features of a disease, and determining feature values of diseases with the frequencies of co-occurrence of two entities. The disease–disease similarities from the proposed method can be used in computational biology researches which use similarities among diseases.

Introduction

Diseases are functionally connected to one another. One gene can cause various diseases, or inhibiting protein translation by one miRNA can be a contributing factor to various diseases. Therefore, a person with a certain disease has a higher probability of getting functionally connected disease than normal people. By using the disease connection information, the possibility of a specific disease onset for a person can be predicted and it is a simple example showing how disease–disease similarity can be utilized for disease-related function research. Disease–disease similarity will be of much help to disease research. It can be useful for development of new drug by aiding in drug repositioning, for searching new genes related to disease, and it can increase efficiency of network analysis in disease-related function research by enhancing disease networks.

There are three primary approaches to get disease–disease similarity: function-based approaches [1], [2], [3] and semantic-based approaches [4], [5], [6], [7], [8], [9], [10], [11], [12], and hybrid approaches of combining previous two approaches [13]. To seek the disease–disease similarity, function-based approaches compare functionally related genes, pathways and biological processes, and semantic-based approaches find similarity between disease terms of ontology related to diseases. Hybrid approaches utilize both functional similarity and semantic similarity. Liu et al. [1] calculated disease–disease similarity using both genetic information from GAD (Genetic Association Database) and environmental etiological factors from MeSH (Medical Subject Headings). Suthram et al. [2] calculated disease–disease similarity using mRNA expression from GEO (Gene Expression Omnibus) database and protein–protein interaction from HPRD (Human Protein Reference Database). Mathur and Dinakarpandian [3] calculated disease–disease similarity using semantic similarity of biological process based on gene ontology. In Li's case [4], a software package for calculating disease–disease similarity was developed using semantic similarities among terms of disease ontology and in the software, 10 methods of seeking semantic similarity are applied to disease ontology in calculating disease–disease similarity. Lastly, Cheng's [13] is a hybrid approach, which first calculates association score utilizing a gene function network and disease-related gene set, and secondly calculates semantic score on disease ontology, and finally gets disease–disease similarity adding these two scores.

Many existing methods find disease–disease similarity using genetic information or semantic information on gene ontology but there are also other similar approaches. Goh et al. [14] constructed human disease network with gene–disease associations from OMIM (Online Mendelian Inheritance in Man) database. They made a connection between two diseases if the diseases shared at least one gene. Lee et al. [15] constructed bipartite human disease association network using shared metabolic pathways from KEGG (Kyoto Encyclopedia of Genes and Genomes) database. Two diseases are linked if mutated enzymes associated with them catalyze adjacent metabolic reactions. Goh's method and Lee's method are related to the proposed method, but they cannot calculate disease-disease similarity. Zhang et al. [16] created feature vectors for disease phenotypes by utilizing phenotype records and calculated cosine similarities among disease phenotypes using the feature vectors, and then developed a disease phenotype network. van Driel et al. [17] employed text mining approach to calculate similarities between diseases. MeSH terms were served as features, and the number of times the term was found in an OMIM record was counted for feature value. They used MeSH hierarchy and inverse document frequency measure to refine feature values. Lastly, similarity between two diseases was computed by the cosine of the angle between their corresponding feature vectors. Hamaneh et al. [18] calculated disease–disease similarity by considering information flow on disease-protein network. The disease–protein network was made by using disease–gene associations from CTD (The Comparative Toxicogenomics Database) database and protein–protein interactions from ppiTrim database. Proteins were treated as features of a disease, and feature value was defined by the expected number of visits by random walker on the disease–protein network. Then disease–disease similarity was calculated by the cosine of the angle similar to van Driel's method.

Biomedical term relations from literature (research papers) can also help calculate disease–disease similarity. We propose a new literature-based method LDDSim (Literature-Driven Disease Similarities) to measure disease similarity. The proposed method extracts disease–gene relations and disease–drug relations from literature, and with the number of those relations, it builds disease–gene matrix and disease–drug matrix. Then the method calculates disease–disease similarity utilizing mutual information between the two diseases. In addition to it, we constructed disease network using the disease similarities.