Deep Learning Prediction of Antimicrobial Peptides in Protozoa

Abstract

Antimicrobial resistance has become a major global health challenge, creating an urgent need for novel anti-infective molecules. Antimicrobial peptides (AMPs) constitute a promising alternative to conventional antibiotics because of their broad-spectrum antimicrobial activity and immunomodulatory properties. Despite the remarkable diversity of protozoa, very few protozoan AMPs have been experimentally identified. Here, we employed the deep learning algorithm AMPlify to systematically screen the proteomes of Blastocystis hominis, Trypanosoma cruzi, Leishmania major, Toxoplasma gondii, and Entamoeba histolytica for high-confidence AMP candidates. Candidate proteins with prediction probabilities greater than 0.90 were further annotated by sequence similarity searches and prioritized for experimental validation. We identified 48 candidate AMPs in B. hominis, 27 in T. cruzi, 23 in T. gondii, and 24 in E. histolytica, whereas no high-confidence candidates were detected in L. major. Predicted candidates represented diverse functional classes, including ribosomal proteins, enzymes, membrane-associated proteins, and proteins involved in nucleic acid metabolism. Three peptides were synthesized and evaluated against Escherichia coli K12. Among them, a peptide derived from the 60S ribosomal protein L3 of B. hominis inhibited bacterial growth with a minimum inhibitory concentration of 125 µM, whereas peptides derived from ribosomal protein L8 and T. cruzi trans-sialidase showed no detectable activity. These findings demonstrate the utility of deep learning for prioritizing novel AMP candidates while highlighting the importance of experimental validation. The identified candidates provide a valuable resource for future functional studies and the development of peptide-based antimicrobial therapeutics.