Limits...
Leveraging 3D chemical similarity, target and phenotypic data in the identification of drug-protein and drug-adverse effect associations.

Vilar S, Hripcsak G - J Cheminform (2016)

Bottom Line: Drug-target identification is crucial to discover novel applications for existing drugs and provide more insights about mechanisms of biological actions, such as adverse drug effects (ADEs).Both modules were linked to generate a final predictor that establishes hypothesis about new drug-target-adverse effect candidates.The modeling described in the current study is simple and efficient and has applications at large scale in drug repurposing and drug safety through the identification of mechanism of action of biological effects.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Informatics, Columbia University Medical Center, New York, NY USA.

ABSTRACT

Background: Drug-target identification is crucial to discover novel applications for existing drugs and provide more insights about mechanisms of biological actions, such as adverse drug effects (ADEs). Computational methods along with the integration of current big data sources provide a useful framework for drug-target and drug-adverse effect discovery.

Results: In this article, we propose a method based on the integration of 3D chemical similarity, target and adverse effect data to generate a drug-target-adverse effect predictor along with a simple leveraging system to improve identification of drug-targets and drug-adverse effects. In the first step, we generated a system for multiple drug-target identification based on the application of 3D drug similarity into a large target dataset extracted from the ChEMBL. Next, we developed a target-adverse effect predictor combining targets from ChEMBL with phenotypic information provided by SIDER data source. Both modules were linked to generate a final predictor that establishes hypothesis about new drug-target-adverse effect candidates. Additionally, we showed that leveraging drug-target candidates with phenotypic data is very useful to improve the identification of drug-targets. The integration of phenotypic data into drug-target candidates yielded up to twofold precision improvement. In the opposite direction, leveraging drug-phenotype candidates with target data also yielded a significant enhancement in the performance.

Conclusions: The modeling described in the current study is simple and efficient and has applications at large scale in drug repurposing and drug safety through the identification of mechanism of action of biological effects.

No MeSH data available.


Related in: MedlinePlus

ROC curve (a) for the global drug-target predictor along with precision (b) and enrichment factors (c) in different top positions
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC4930585&req=5

Fig2: ROC curve (a) for the global drug-target predictor along with precision (b) and enrichment factors (c) in different top positions

Mentions: We integrated 3D chemical/pharmacophoric similarity into target data from ChEMBL [36] as described in Methods (1526 drugs and 726 targets). Our predictor generated 1,107,876 drug-target combinations with associated leave-one-out scores. Each drug-target candidate is associated with the 3D maximum similarity score against the set of drugs that interact with the same target according to ChEMBL. We labeled as true positives (TP) the drug-target associations already collected in ChEMBL and as false positives (FP) the rest of possible combinations (we defined the FP cases from the unknown cases with no target information collected in the ChEMBL). ROC curve was plotted with an area of 0.82 (see Fig. 2a). We also plotted precision and enrichment factor (EF) in different top positions for the global drug-target predictor (see Fig. 2b, c).Fig. 2


Leveraging 3D chemical similarity, target and phenotypic data in the identification of drug-protein and drug-adverse effect associations.

Vilar S, Hripcsak G - J Cheminform (2016)

ROC curve (a) for the global drug-target predictor along with precision (b) and enrichment factors (c) in different top positions
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC4930585&req=5

Fig2: ROC curve (a) for the global drug-target predictor along with precision (b) and enrichment factors (c) in different top positions
Mentions: We integrated 3D chemical/pharmacophoric similarity into target data from ChEMBL [36] as described in Methods (1526 drugs and 726 targets). Our predictor generated 1,107,876 drug-target combinations with associated leave-one-out scores. Each drug-target candidate is associated with the 3D maximum similarity score against the set of drugs that interact with the same target according to ChEMBL. We labeled as true positives (TP) the drug-target associations already collected in ChEMBL and as false positives (FP) the rest of possible combinations (we defined the FP cases from the unknown cases with no target information collected in the ChEMBL). ROC curve was plotted with an area of 0.82 (see Fig. 2a). We also plotted precision and enrichment factor (EF) in different top positions for the global drug-target predictor (see Fig. 2b, c).Fig. 2

Bottom Line: Drug-target identification is crucial to discover novel applications for existing drugs and provide more insights about mechanisms of biological actions, such as adverse drug effects (ADEs).Both modules were linked to generate a final predictor that establishes hypothesis about new drug-target-adverse effect candidates.The modeling described in the current study is simple and efficient and has applications at large scale in drug repurposing and drug safety through the identification of mechanism of action of biological effects.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Informatics, Columbia University Medical Center, New York, NY USA.

ABSTRACT

Background: Drug-target identification is crucial to discover novel applications for existing drugs and provide more insights about mechanisms of biological actions, such as adverse drug effects (ADEs). Computational methods along with the integration of current big data sources provide a useful framework for drug-target and drug-adverse effect discovery.

Results: In this article, we propose a method based on the integration of 3D chemical similarity, target and adverse effect data to generate a drug-target-adverse effect predictor along with a simple leveraging system to improve identification of drug-targets and drug-adverse effects. In the first step, we generated a system for multiple drug-target identification based on the application of 3D drug similarity into a large target dataset extracted from the ChEMBL. Next, we developed a target-adverse effect predictor combining targets from ChEMBL with phenotypic information provided by SIDER data source. Both modules were linked to generate a final predictor that establishes hypothesis about new drug-target-adverse effect candidates. Additionally, we showed that leveraging drug-target candidates with phenotypic data is very useful to improve the identification of drug-targets. The integration of phenotypic data into drug-target candidates yielded up to twofold precision improvement. In the opposite direction, leveraging drug-phenotype candidates with target data also yielded a significant enhancement in the performance.

Conclusions: The modeling described in the current study is simple and efficient and has applications at large scale in drug repurposing and drug safety through the identification of mechanism of action of biological effects.

No MeSH data available.


Related in: MedlinePlus