Limits...
Concept Modeling-based Drug Repositioning.

Patchala J, Jegga AG - AMIA Jt Summits Transl Sci Proc (2015)

Bottom Line: To test this, we constructed a probabilistic topic model based on the Unified Medical Language System (UMLS) concepts that appear in the disease and drug related abstracts in MEDLINE.The resulting probabilistic topic associations were used to measure the similarity between disease and drugs.The success of the proposed model is evaluated using a set of repositioned drugs, and comparing a drug's ranking based on its similarity to the original and new indication.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science, University of Cincinnati, Cincinnati, Ohio, USA.

ABSTRACT
Our hypothesis is that drugs and diseases sharing similar biomedical and genomic concepts are likely to be related, and thus repositioning opportunities can be identified by ranking drugs based on the incidence of shared similar concepts with diseases and vice versa. To test this, we constructed a probabilistic topic model based on the Unified Medical Language System (UMLS) concepts that appear in the disease and drug related abstracts in MEDLINE. The resulting probabilistic topic associations were used to measure the similarity between disease and drugs. The success of the proposed model is evaluated using a set of repositioned drugs, and comparing a drug's ranking based on its similarity to the original and new indication. We then applied the model to rare disorders and compared them to all approved drugs to facilitate "systematically serendipitous" discovery of relationships between rare diseases and existing drugs, some of which could be potential repositioning candidates.

No MeSH data available.


Related in: MedlinePlus

Schematic representation of overall workflow. Drug and disease-related abstracts are Metamapped to generate a list of biomedical and genomic CUIs from UMLS for each drug and disease. Topic modeling is then applied followed by statistical analysis to assess the similarity between disease and drug.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4525261&req=5

f1-2089413: Schematic representation of overall workflow. Drug and disease-related abstracts are Metamapped to generate a list of biomedical and genomic CUIs from UMLS for each drug and disease. Topic modeling is then applied followed by statistical analysis to assess the similarity between disease and drug.

Mentions: The topic model is a state-of-the-art Bayesian model for extracting semantic structure from document collections9. It automatically learns a set of thematic topics (lists of words or “bag of words”) that describe a document collection, and assigns the topics to each of the documents in the collection with a probability value. Topic models have recently retained a lot of attention and have been used to address various issues (e.g., drug repositioning10, word sense disambiguation in the clinical domain11, gene-drug relationship extraction from literature12, etc.). As a variation of classic “bag-of-words” approach, we use a “bag of concepts” approach. We first employ the UMLS Metathesaurus to identify biomedical concepts and construct a probabilistic topic model based on the concepts that appear in the disease and drug related abstracts. The resulting probabilistic topic associations are used to measure the similarity between disease and drugs and identify drug repositioning candidates (Fig. 1).


Concept Modeling-based Drug Repositioning.

Patchala J, Jegga AG - AMIA Jt Summits Transl Sci Proc (2015)

Schematic representation of overall workflow. Drug and disease-related abstracts are Metamapped to generate a list of biomedical and genomic CUIs from UMLS for each drug and disease. Topic modeling is then applied followed by statistical analysis to assess the similarity between disease and drug.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4525261&req=5

f1-2089413: Schematic representation of overall workflow. Drug and disease-related abstracts are Metamapped to generate a list of biomedical and genomic CUIs from UMLS for each drug and disease. Topic modeling is then applied followed by statistical analysis to assess the similarity between disease and drug.
Mentions: The topic model is a state-of-the-art Bayesian model for extracting semantic structure from document collections9. It automatically learns a set of thematic topics (lists of words or “bag of words”) that describe a document collection, and assigns the topics to each of the documents in the collection with a probability value. Topic models have recently retained a lot of attention and have been used to address various issues (e.g., drug repositioning10, word sense disambiguation in the clinical domain11, gene-drug relationship extraction from literature12, etc.). As a variation of classic “bag-of-words” approach, we use a “bag of concepts” approach. We first employ the UMLS Metathesaurus to identify biomedical concepts and construct a probabilistic topic model based on the concepts that appear in the disease and drug related abstracts. The resulting probabilistic topic associations are used to measure the similarity between disease and drugs and identify drug repositioning candidates (Fig. 1).

Bottom Line: To test this, we constructed a probabilistic topic model based on the Unified Medical Language System (UMLS) concepts that appear in the disease and drug related abstracts in MEDLINE.The resulting probabilistic topic associations were used to measure the similarity between disease and drugs.The success of the proposed model is evaluated using a set of repositioned drugs, and comparing a drug's ranking based on its similarity to the original and new indication.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science, University of Cincinnati, Cincinnati, Ohio, USA.

ABSTRACT
Our hypothesis is that drugs and diseases sharing similar biomedical and genomic concepts are likely to be related, and thus repositioning opportunities can be identified by ranking drugs based on the incidence of shared similar concepts with diseases and vice versa. To test this, we constructed a probabilistic topic model based on the Unified Medical Language System (UMLS) concepts that appear in the disease and drug related abstracts in MEDLINE. The resulting probabilistic topic associations were used to measure the similarity between disease and drugs. The success of the proposed model is evaluated using a set of repositioned drugs, and comparing a drug's ranking based on its similarity to the original and new indication. We then applied the model to rare disorders and compared them to all approved drugs to facilitate "systematically serendipitous" discovery of relationships between rare diseases and existing drugs, some of which could be potential repositioning candidates.

No MeSH data available.


Related in: MedlinePlus