Unraveling the evolutionary logic and structure of the marine resistome: a deep structured learning approach to discovery of evolutionary constraints

• Grant number: 21/08235-3
• Support Opportunities: Scholarships in Brazil - Doctorate
• Effective date (Start): July 01, 2022
• Effective date (End): February 29, 2024
• Field of knowledge: Health Sciences - Pharmacy
• Principal Investigator: Ricardo Roberto da Silva
• Grantee: Tiago Cabral Borelli
• Host Institution: Faculdade de Ciências Farmacêuticas de Ribeirão Preto (FCFRP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil
• Associated research grant: 17/18922-2 - Development of a computing platform extensible and modular for metabolomics and metagenomics analysis: innovation with the discovery of new enzymatic activities and natural products of pharmaceutical interest derived, AP.BTA.JP

Abstract

The increasing rates of antimicrobial resistance have prompted the use of new approaches to prevent or mitigate resistance in pathogens. Historically, soil microbial communities or impacted marine environments were sources of new antimicrobials. Recently, data obtained from preserved marine environments have enabled the discovery of distant homologous genes for the production of novel antimicrobials, which puts this environment under selective pressure. Therefore, it is prone to evolution of antibiotic resistance genes. In addition, laboratory evolution studies have shown constraints and patterns on antimicrobial resistance evolution, since cells have to preserve essential functions for growth and multiplication. This suggests that evolutionary pathways can be targeted and predicted. For some resistance mechanisms, such as genes encoding enzymes for degradation of the antimicrobial compound or drug efflux transporters, the acquisition of resistance can be easily understood. However, the effects of genetic mutations (Single Nucleotide Polymorphisms, SNPs) or horizontal transfers are not simply additive, and complex genetic interactions between mutations and unknown mechanisms are often encountered. Data-derived analysis approaches, such as machine learning, can be used to capture the nonlinear relationships between observed mutations and resistance phenotypes. Machine learning models, in particular deep learning, have been used to predict resistance phenotypes based on genetic polymorphisms, as well as to predict the antimicrobial action of chemical compounds. However, these approaches are limited to predicting resistance or susceptibility to discrete antimicrobial actions, without necessarily understanding the evolutionary pathways that explain these phenotypes. Also, most resistance gene prediction strategies are still based on best hits on database sequence similarity searches, which suffer from low annotation coverage and result in a high rate of false negatives. At the same time, antimicrobial substances and genes conferring resistance mechanisms are not properly categorized and their evolutionary constraints are not explored. Thus, this project aims to expand and structure resistance genes bases such as ARDB, CARD and DeepARG-DB and the creation of multi-class classifiers using deep structured learning to predict resistance genes in data from metagenomics of microbial communities in marine environments and identifying evolutionary constraints on antimicrobial resistance. Data from marine communities will be obtained in the context of the thematic project Integrated approach to sustainable prospecting for natural marine products: from diversity to anticancer substances (Process 2015/17177-6). In parallel, the prediction of antimicrobial activity of large sets of chemical structures, such as PubChem and ZINC15 (> 100 million structures) will also be performed by deep structured learning models, using the infrastructure of the research aid Development of an extensible modular computational platform for analysis of metabolomics and metagenomics experiments (Process: 17/18922-2). This parallel approach will allow a better categorization of resistance mechanism and evolutionary constraints, which will enable the generation of probabilistic models for selection of compounds that are more likely to evade resistance, in addition to identifying genes or gene islands that are more likely to be adapting to the selective pressure imposed by antimicrobials. (AU)