Compensating for literature annotation bias when predicting novel drug-disease relationships through Medical Subject Heading Over-representation Profile (MeSHOP) similarity

Abstract

BackgroundUsing annotations to the articles in MEDLINE®/PubMed®, over six thousand chemical compounds with pharmacological actions have been tracked since 1996. Medical Subject Heading Over-representation Profiles (MeSHOPs) quantitatively leverage the literature associated with biological entities such as diseases or drugs, providing the opportunity to reposition known compounds towards novel disease applications.MethodsA MeSHOP is constructed by counting the number of times each medical subject term is assigned to an entity-related research publication in the MEDLINE database and calculating the significance of the count by comparing against the count of the term in a background set of publications. Based on the expectation that drugs suitable for treatment of a disease (or disease symptom) will have similar annotation properties to the disease, we successfully predict drug-disease associations by comparing MeSHOPs of diseases and drugs.ResultsThe MeSHOP comparison approach delivers an 11% improvement over bibliometric baselines. However, novel drug-disease associations are observed to be biased towards drugs and diseases with more publications. To account for the annotation biases, a correction procedure is introduced and evaluated.ConclusionsBy explicitly accounting for the annotation bias, unexpectedly similar drug-disease pairs are highlighted as candidates for drug repositioning research. MeSHOPs are shown to provide a literature-supported perspective for discovery of new links between drugs and diseases based on pre-existing knowledge.

FLAGS, frequently mutated genes in public exomes

“…Our expectation is that if the frequently seen genes are arising as candidates in more studies, and are less likely to be truly pathogenic, then they could be associated to a wider range of phenotypes in the literature (we recognize the association could also be due to pleiotropy [ 45 ], see Limitations). To analyze if FLAGS have been frequently correlated to human diseases, we used two different computational resources (MeSHOP [ 32 ], HPO [ 33 ]) to extract known significant relationship(s) between genes and human disease phenotypes based on published scientific articles. Figures 5 a and b show the distribution of the number of disease terms from HPO and MeSHOP per gene within gene sets.…”

“…We used MeSHOP software [ 32 ] to identify over-represented disease terms associated with each gene. MeSHOP returns a list of MeSH (Medical Subject Heading) terms for each gene with a p-value for each term.…”

Dynamic software design for clinical exome and genome analyses: insights from bioinformaticians, clinical geneticists, and genetic counselors

“…Clinicians emphasized that while there are tools that offer online software applications to obtain candidate genes based upon keyword queries (e.g., MeSHOP, 34 Genie, 35 Ingenuity [ http://www.ingenuity.com ]), these capabilities are not consistently accessible to integrated WES/WGS analysis software and the output cannot be combined with exome data without additional manipulation. Expanding beyond keywords as input, clinicians further requested graphical search functionalities.…”

‘RE:fine drugs’: an interactive dashboard to access drug repurposing opportunities

“…Thus, we test the effectiveness of our framework to enrich drug reposition candidates using a permutation test. We use the co-occurrence of a drug and a disease in biomedical literature and clinical trials database as a proxy of biological plausibility of the drug reposition, since it can be influenced by literature bias ( 20 , 21 ). Our study shows that 28% (14 816 out of 52 966) of the transitive drug–disease pairs from PheWAS were supported in at least one Medline abstract while permuting drug–disease pairs will only identify 6.2% of random drug–disease words in the literature.…”

“…Furthermore, to estimate the novelty of the drug–disease pairs and to prioritize them, we cross referenced all pairs with the NIH clinical trial registry (i) and also used the biomedical literature co-occurrences of drug–disease pairs (ii) to categorize each PheWAS and GWAS generated pair in one of these four categories: ‘known/rediscovered’ for relationships that already exist in the DrugBank, ‘strongly supported’ for pairs with some evidence in both (i) and (ii), ‘Likely’ for pair with support in either (i) or (ii) and ‘novel’ for pairs with no evidence in (i) or (ii). Note that the lack of occurrence of both drug and disease terms together in the literature could potentially indicate a novel finding, a bias of literature annotations or a true lack of association ( 20 , 21 ).…”