Authors:
Eduardo Campos dos Santos
1
;
Braulio Roberto Gonçalves Marinho Couto
2
;
Marcos A. dos Santos
3
and
Julio Cesar Dias Lopes
3
Affiliations:
1
Instituto de Ciências Biológicas and Universidade Federal de Minas Gerais / UFMG, Brazil
;
2
Centro Universitário de Belo Horizonte / UNI-BH, Brazil
;
3
UFMG, Brazil
Keyword(s):
Human drug target, Logistic regression, Case-control study, Prediction models.
Related
Ontology
Subjects/Areas/Topics:
Algorithms and Software Tools
;
Bioinformatics
;
Biomedical Engineering
;
Pharmaceutical Applications
Abstract:
Drug target identification and validation are critical steps in the drug discovery pipeline. Hence, predicting potential “druggable targets”, or targets that can be modulated by some drug, is very relevant to drug discovery. Approaches using structural bioinformatics to predict “druggable domains” have been proposed, but they have only been applied to proteins that have solved structures or that have a reliable model predicted by homology. We show that available protein annotation terms may be used to explore semantic-based measures to provide target similarity searching and develop a tool for potential drug target prediction. We analysed 1,541 human protein drug targets and 29,580 human proteins not validated as drug targets but which share some InterPro annotations with a known drug target. We developed a semantic-based similarity measure by using singular value decomposition over InterPro terms associated with drug targets, performed statistical analyses and built logistic regress
ion models. We present a probabilistic model summarised in a closed mathematical formula that allows human protein drug targets to be predicted with a sensitivity of 89% and a specificity of 67%.
(More)