In silico analysis of the antimicrobial activity of phytochemicals: towards a technological breakthrough

Highlights

• Experimental use of in silico instruments, feedforward Multi-Layer Perceptron, and Genetic Programming, to predict the effectiveness of natural and experimental mixtures of polyphenols against several microbial strains.

• Multi-Layer Perceptron shows high correlation in predicting the antimicrobial sensitivity.

• Genetic Programming gives explicit representation of the acquired knowledge about the polyphenols.

• In silico analysis provides a useful strategy to innovate the classic microbiological assays.

Abstract

Background

The complications associated with infections from pathogens increasingly resistant to traditional drugs lead to a constant increase in the mortality rate among those affected. In such cases the fundamental purpose of the microbiology laboratory is to determine the sensitivity profile of pathogens to antimicrobial agents. This is an intense and complex work often not facilitated by the test's characteristics. Despite the evolution of the Antimicrobial Susceptibility Testing (AST) technologies, the technological breakthrough that could guide and facilitate the search for new antimicrobial agents is still missing.

Methods

In this work, we propose the experimental use of in silico instruments, particularly feedforward Multi-Layer Perceptron (MLP) Artificial Neural Network, and Genetic Programming (GP), to verify, but also to predict, the effectiveness of natural and experimental mixtures of polyphenols against several microbial strains.

Results

We value the results in predicting the antimicrobial sensitivity profile from the mixture data. Trained MLP shows very high correlations coefficients (0,93 and 0,97) and mean absolute errors (110,70 and 56,60) in determining the Minimum Inhibitory Concentration and Minimum Microbicidal Concentration, respectively, while GP not only evidences very high correlation coefficients (0,89 and 0,96) and low mean absolute errors (6,99 and 5,60) in the same tasks, but also gives an explicit representation of the acquired knowledge about the polyphenol mixtures.

Conclusions

In silico tools can help to predict phytobiotics antimicrobial efficacy, providing an useful strategy to innovate and speed up the extant classic microbiological techniques.

Introduction

Microbes are infectious agents, widely spread in nature. Some of them are responsible for diseases in humans (pathogens). They are divided into: bacteria, fungi, viruses, prions and protozoa. They also include macroscopic parasites. The microbiology laboratory plays an essential role in pathogen infection prevention and control and should be able to identify microbes responsible for infections [88].

Once the pathogens have been identified, the fundamental purpose of the microbiology laboratory is then to determine the sensitivity profile of pathogens to antimicrobial agents. Antibiogram analysis for the definition of antimicrobial susceptibility plays a basic role in microbiological diagnostic and the research of new effective antimicrobial agents. Despite the evolution and innovation of the microbiological techniques, the culture-based (phenotypic) tests, including disc diffusion and broth dilution methods, remain the gold standard methods to analyze microorganism susceptibility/resistance to antimicrobial drugs. In particular, the easy and low-cost disc diffusion method is commonly used in daily practice and is considered the most widespread in the world [9,88]. Also, the Antimicrobial Susceptibility Testing (AST) remains essential for the characterization of isolates for taxonomic, epidemiological and clinical purposes, for the definition of the Antimicrobial Resistance (AMR) and the study of new antimicrobial agents activity [84].

Some bioinformatic tools that predict AMR and/or AST from genomic sequence data have been developed [30,90,91]. Machine learning algorithms [32] represent an alternative method to predict AMR, for various bacterial species and antibiotic combinations, from sequence data [14,23,28,53,57]. Furthermore, a neural network has been recently developed for AMR prediction of multiple bacterial species [7].

Currently, the diffusion and dilution methods are the most employed bioassays to analyze the antimicrobial properties of natural botanical compounds or phytochemicals. However, such methods are not suitable to standardise the study process of natural antimicrobial extracts. Indeed, a series of factors (such as culture medium composition, tested microorganisms, botanical features, geographical origin of plant matrix, extracting method, pH, the solubility of the extracts in the culture media, and others) are responsible for high variability, as underlined by several data examined in past and recent reviews and original articles [[8], [49], [71]].

Phytochemicals represent relevant molecules with a high economic value [27]. Historically, natural products have accounted for more than one-half of all therapeutic agents and, today, natural products are the inspiration for almost 40% of new drugs on the market [25]. The spectrum of activity of the natural compounds as therapeutic molecules is wide, ranging from antioxidant, immunostimulant, antimicrobial, antiviral, antifungal, anticancer, anti-inflammatory activities [55]. In the last years, the scientific study of the biological activity of natural products has been very intense, showing that there is a close connection between the chemical structure of the natural molecule (number, position and type of functional groups, the structure of the carbon skeleton, length and type of the side chains, saturation degree, etc.) and the biological activity [63]. However, predicting the biological activity basing on the analysis of the chemical components of a mixture appears impossible. A mixture is composed of dozens or even hundreds of chemical agents that act synergistically [16]. Moreover, no analysis allows to evaluate exactly the nature and the amount of all the chemical compounds present in a mixture. The complete characterization of secondary metabolites chemical profile is almost impossible and each technique (GC-MS, LC-MS, HPLC, NMR, etc.) has got limitations and drawbacks [75]. In a mixture, the presence of secondary metabolites in very small quantities, therefore difficult to characterize, makes the natural mixtures extremely variable both in composition and biological activity.

Due to the high number of variables that can influence the biological activity of individual or mixtures of compounds, it is necessary to conduct a high number of screening tests to verify the cytotoxic power, the biological action type (antibacterial, antioxidant, antitumor, etc.) and the active concentration. It woud be therefore desirable to develop an in silico method that allows obtaining information before conducting the biological tests, to reduce the analysis times and costs.

In this paper, some previously published experimental results obtained in in vitro assays on the antimicrobial effects of different polyphenols (a class of phytochemicals) have been analyzed with computer skills, in an attempt to evaluate the capacity of in silico instruments to highlight the correlation between the characteristics of the extracts and the antimicrobial efficacy. Namely, we apply a feedforward Multi-Layer Perceptron (MLP) Artificial Neural Network (ANN) [[10], [11], [38]], and a Genetic Programming (GP) procedure [19,35,43,44] to generate and evolve automatically unknown functions represented implicitally (ANN) or explicitally (GP).

While these methods have already been applied in many areas of life sciences [20,21,48,[66], [67], [68], [69],82,83], they never seem to have been used to predict an antimicrobial sensitivity profile using an approach not based on the culture of the microorganism and on the data of the genomic sequence.