Novel Antimicrobials from Computational Modelling and Drug Repositioning: Potential In Silico Strategies to Increase Therapeutic Arsenal Against Antimicrobial Resistance

Abstract

Antimicrobial resistance (AMR) is one of the most significant public health threats today. The need for new antimicrobials against multidrug-resistant infections is growing. The development of computational models capable of predicting new drug-target interactions is an interesting strategy to reposition already known drugs into potential antimicrobials. The objective of this review was to compile the latest advances in the development of computational models capable of identifying drugs already registered by the Food and Drug Administration for other indications with potential capacity to be applied as antimicrobials. We present studies that apply in silico methods such as machine learning, molecular docking, molecular dynamics and deep learning. Some of these studies have in vitro/in vivo results that demonstrate the reliability of this computational methodology in terms of the identification of effective molecules and new targets of interest in the treatment of infections. In addition, we present the methods that are under development and their future prospects in terms of the search for new antimicrobials. We highlight the need to implement these strategies in the research of effective drugs in the treatment of infectious diseases and to continue to improve the available models and approaches to gain an advantage against the rapid emergence of AMR.

Keywords: QSAR models; antimicrobial resistance; computational models; deep learning; drug repositioning; machine learning; mathematical prediction models; molecular docking; molecular dynamics; topological data analysis.

Figure 1

Duration of drug development from de novo synthesis compared to the use of computational methods and drug repositioning.

Figure 2

The main computational strategies for the discovery of new antimicrobials and their combinations. The combinations have been classified into 3 categories: the widely applied ones (solid line) are the most used because they are employed separately, and their results have a known criterion; the frequently applied ones (large dashed line), although they do not present as great a criterion as the previous ones, when combined, provide information that cannot be obtained with the separate methods; and the occasionally applied ones (small dashed line) are carried out for specific questions or needs of the study to be carried out (genomic data of interest or molecular structures that cannot be observed with more traditional computational models); however, they are young computational models with limited reliability.

Figure 3

Graphical scheme of the steps commonly followed in the subtractive genomics approach. (A): Retrieval of interest and reference genomes from databases. (B): Removal of paralogous sequences in the genome of interest (GI). (C): Removal of homologous sequences to the reference genome (RG). (D): Compare non-homologous against database of essential genes to identify essential genes. (E): Identify and test for high similarity of non-homologous essential proteins to those found in databases of known targets. (F): Selection of targets of interest based on criteria of virulence, resistance, antigenicity, localization and function in the pathogen.