Accelerating antibiotic discovery from Earth's isolated habitats through protein language models and microfluidics

MicroSPEAR identified 85 synthesized antimicrobial peptide candidates, with 79 showing antibacterial activity at ≤64 M, and four broad-spectrum, low-toxicity leads reducing bacterial burdens in a mouse thigh infection model.

• Why it matters: Earth’s isolated habitat microbiomes are largely untapped for antimicrobial peptides, but divergent peptide sequences and slow testing make reliable prioritization and validation difficult.
• What they did: The framework couples SPEAR, a sequence–structure-aware ensemble predictor that generalized to low-homology peptides, with MicroGrad microfluidic susceptibility testing that reduced assay time from ~18 h to 4 h.
• The result: Across a catalogue from 4,480 metagenomes spanning 15 isolated habitat types, SPEAR produced 383,027 non-redundant candidate AMPs, 63.1% lacking detectable homologues, and mechanistic analyses linked candidates to distinct membrane disruption modes.

AI summary of the abstract · may contain errors · not medical advice · read the paper.