Identification of Novel Antimicrobial Resistance Genes Using Machine Learning, Homology Modeling, and Molecular Docking

Abstract

Antimicrobial resistance (AMR) threatens the healthcare system worldwide with the rise of emerging drug resistant infectious agents. AMR may render the current therapeutics ineffective or diminish their efficacy, and its rapid dissemination can have unmitigated health and socioeconomic consequences. Just like with many other health problems, recent computational advances including developments in machine learning or artificial intelligence hold a prodigious promise in deciphering genetic factors underlying emergence and dissemination of AMR and in aiding development of therapeutics for more efficient AMR solutions. Current machine learning frameworks focus mainly on known AMR genes and are, therefore, prone to missing genes that have not been implicated in resistance yet, including many uncharacterized genes whose functions have not yet been elucidated. Furthermore, new resistance traits may evolve from these genes leading to the rise of superbugs, and therefore, these genes need to be characterized. To infer novel resistance genes, we used complete gene sets of several bacterial strains known to be susceptible or resistant to specific drugs and associated phenotypic information within a machine learning framework that enabled prioritizing genes potentially involved in resistance. Further, homology modeling of proteins encoded by prioritized genes and subsequent molecular docking studies indicated stable interactions between these proteins and the antimicrobials that the strains containing these proteins are known to be resistant to. Our study highlights the capability of a machine learning framework to uncover novel genes that have not yet been implicated in resistance to any antimicrobials and thus could spur further studies targeted at neutralizing AMR.

Antibiotic Resistance in India: Counteractive Consequences Using Antibiotics During the COVID-19 Pandemic

“…Given the current situation in combating AMR, innovative methods to pinpoint and priorities the genetic causes of AMR that are still unidentified are urgently required [12]. To mitigate bacterial resistance to antibiotics, novel antimicrobial methods and medications are therefore urgently needed [11].…”

Tackling the Antimicrobial Resistance “Pandemic” with Machine Learning Tools: A Summary of Available Evidence

“…A reservoir of resistance genes may come from previously unknown sources. Other than identifying known resistance genes, machine learning algorithms may be utilised to identify novel yet unrecognised resistance genes [40]. Artificial intelligence has the potential to predict the presence and spread of antimicrobial resistance genes within and across populations [4].…”

Silicon versus Superbug: Assessing Machine Learning’s Role in the Fight against Antimicrobial Resistance

“…Using bacterial isolates filtered by genotypic and AST phenotypic data from the NCBI Pathogen Detection database, the researchers constructed an AMR-AST matrix for each combination of antibiotic-species groupings. This matrix consisted of the features (genes), binary AST labels as target classes, and sample accession numbers to be input into a variety of ML algorithms to perform supervised binary classification [64]. These algorithms were trained and tested using 6-fold stratified cross-validation-implemented in StratifiedKFold-for all genes and AST phenotype data to ensure that genes deemed important for discrimination were obtained.…”