Mustafa Alhaji Isa¹,4, Ibrahim Abubakar Mustapha2 and Tijjani Isa Bulamarwa3
¹Department of Biological Sciences, Kashim Ibrahim University, Maiduguri, Borno State, Nigeria
2Department of Islamic Studies, Kashim Ibrahim University, Maiduguri, Borno State, Nigeria
3Department of Physic, Kashim Ibrahim University, Maiduguri, Borno State, Nigeria
4Department of Microbiology, University of Maiduguri, Maiduguri, Nigeria
*Corresponding author’s Email: mustafaalhajiisa@gmail.com, doi.org/10.55639/607.020100154
ABSTRACT
Genomic prediction of antimicrobial resistance (AMR) offers an approach to earlier susceptibility assessment through bacterial genome sequencing, although the transportability of prediction models depends on genetic relatedness, clinical setting, antimicrobial exposure and laboratory practice. This review examines published primary evaluations and benchmark studies of genome-based phenotypic resistance prediction, focusing on internal versus independent validation, isolate provenance and the evidence needed for clinical implementation. Comparisons distinguish random partitions, lineage-aware evaluations, independent collections, temporal tests and laboratory-to-laboratory assessments. In a benchmark covering 11 bacterial species, 44 antimicrobials and 31,195 genomes, the proportion of experiments achieving macro F1 ≥0.9 was 64% under random splitting, 25% under phylogeny-aware splitting and 33% under homology-aware splitting. Individual investigations also describe variation across bacterial species, drugs, independent collections and bioinformatics pipelines. Susceptibility breakpoint definitions, the number of phenotypically resistant isolates, prediction abstention and calibration influence the clinical interpretation of these results. Geographic surveillance evidence further indicates an uneven distribution of publicly shared pathogen genomes, emphasising the importance of recording collection locations separately for development and validation cohorts. The evidence supports population-aware benchmarking, organism-drug-specific clinical error reporting, transparent treatment of uncertain calls and independent evaluation in the intended deployment population. An integrated framework is presented for assessing genomic predictors in relation to biological transportability, geographic representation and clinical utility.
KEYWORDS
Antimicrobial
Resistance,
Whole-Genome
Sequencing,
Machine
Learning,
Susceptibility
Prediction,
External
Validation.