ADDI: Recommending alternatives for drug-drug interactions with negative health effects.
Comput Biol Med
; 125: 103969, 2020 10.
Article
em En
| MEDLINE
| ID: mdl-32836102
ABSTRACT
Investigating the interactions among various drugs is an indispensable issue in the field of computational biology. Scientific literature represents a rich source for the retrieval of knowledge about the interactions between drugs. Predicting drug-drug interaction (DDI) types will help biologists to evade hazardous drug interactions and support them in discovering potential alternatives that increase therapeutic efficacy and reduce toxicity. In this paper, we propose a general-purpose method called ADDI (standing for Alternative Drug-Drug Interaction) that applies deep learning on PubMed abstracts to predict interaction types among drugs. As an application, ADDI recommends alternatives for drug-drug interactions (DDIs) which have Negative Health Effects Types (NHETs). ADDI clearly outperforms state-of-the-art methods, on average by 13%, with respect to accuracy by using only the textual content of the online PubMed papers. Additionally, manual evaluation of ADDI indicates high precision in recommending alternatives for DDIs with NHETs.
Palavras-chave
Texto completo:
1
Base de dados:
MEDLINE
Assunto principal:
Preparações Farmacêuticas
/
Biologia Computacional
Tipo de estudo:
Guideline
Idioma:
En
Revista:
Comput Biol Med
Ano de publicação:
2020
Tipo de documento:
Article
País de afiliação:
Irã