Afri ARM: A curated database of African Antimicrobial Resistance Genes for One Health Surveillance

Abstract

<ns3:p> We present Afri ARM v.1.0, a curated database of ARGs derived from 184 whole-genome shotgun metagenomes from 31 sub-Saharan African countries. The database was built using DIAMOND BLASTX alignment against CARD protein sequences with stringent identity (≥80%) and E-value (≤1×10⁻¹⁰) thresholds. Resistome analysis revealed substantial geographic heterogeneity, with Burkina Faso, Malawi, Benin, Seychelles, and Mali exhibiting the highest normalized ARG intensity. β-lactam resistance dominated drug class annotations, with <ns3:italic>Pseudomonas aeruginosa</ns3:italic> as the leading ARG-harboring taxon. ESM-2 protein language model embeddings and UMAP dimensionality reduction revealed that African ARGs broadly overlap with the global CARD resistome, reflecting globally conserved resistance functions rather than Africa-specific divergence. A supervised random forest classifier trained on 25 physicochemical and amino acid composition features achieved high accuracy in discriminating ARGs from non-ARG proteins. AfriARM provides a reproducible biocuration framework to support One Health AMR surveillance and public health decision-making. </ns3:p>