ngangao/AfriARM

Curation of a database for Antimicrobial Resistance Genes from African metagenomes, protein language modeling, and machine learning for functional landscape analysis # AfriARM Curation of a database for Antimicrobial Resistance Genes from African metagenomes, protein language modeling, and machine learning for functional landscape analysis. A preprint of this work is available here doi.org This repository contains scripts used to curate the Afri-ARM database, a harmonized collection of antimicrobial resistance gene (ARG) detections from African whole-genome shotgun metagenomes. ## Overview Raw metagenomic reads were downloaded from the European Nucleotide Archive (ENA) and uniformly screened against the **Comprehensive Antibiotic Resistance Database (CARD)** using **DIAMOND BLASTX**. All samples were processed using identical thresholds and resource-conservative settings to ensure reproducibility across environments with limited computational capacity. Fasta sequences were used for embeddings and machine learning. ## Workflow 1. Download paired-end metagenomic reads from ENA (SRA accessions) 2. Translate and align reads against CARD protein sequences using DIAMOND 3. Filter alignments by identity and e-value 4. Aggregate ARG hits across samples into a master table 5. Construct a non-redundant FASTA of detected ARG proteins 6. Generate protein embeddings with ESM-2 protein language model and compare to reference ARGs from the Comprehensive Antibiotic Resistance Database (CARD) 7. Implement a protein sequence–based machine learning workflow to distinguish antibiotic resistance gene (ARG) proteins from non-ARG proteins using physicochemical and amino acid composition features ## Key Outputs - `results/afriarm_arg_hits.tsv` Master table of ARG detections across samples - `metadata/afriarm_metadata_clean.tsv` Harmonized sample metadata (country, source, host, site) - `results/afriarm_reference_subset.fasta` CARD protein sequences detected in African metagenomes - `embeddings/africa_embeddings.npy` NumPy array data storing numerical vector representations of ARG proteins from African ARGs for down …