GIHMsC_candidateAMPs

This record provides the candidate antimicrobial peptide (AMP) catalogue and external benchmarking dataset associated with MicroSPEAR, a framework integrating computational prioritization with rapid experimental validation for AMP discovery from Earth's isolated habitats. The deposited datasets comprise: Candidate AMP catalogue: The complete set of 383,027 non-redundant candidate AMPs predicted by SPEAR from the Global Isolated-Habitat Microbiome smORF Catalog (GIHMsC), constructed from 4,480 metagenomes spanning 15 isolated habitat types. These sequences represent a resource for exploring undercharacterized peptide diversity and prioritizing candidates for experimental investigation. External benchmarking dataset: A curated collection of 333 peptides with 1,980 minimum inhibitory concentration (MIC) measurements across six bacterial pathogens, assembled to evaluate model performance in prioritizing peptides with experimentally measured antimicrobial activity. SPEAR integrates ESM2 sequence embeddings with SaProt structure-aware embeddings and combines random forest and histogram-based gradient boosting predictions to generate an AMP score. The source code, including model training and inference workflows, installation instructions and usage examples, is available at https://github.com/HUST-NingKang-Lab/SPEAR. These resources support reproducible benchmarking, comparative analyses and further exploration of AMP diversity in isolated-habitat microbiomes.