Machine learning-guided identification and experimental characterization of a cationic antifungal peptide active...

To evaluate the antifungal activity, selectivity, stability, and cellular responses associated with a machine learning (ML)-identified cationic antimicrobial peptide against Candida.

Using an ML-guided identification strategy, four candidate peptides were selected and synthesized for broad antimicrobial screening against Gram-negative and Gram-positive bacteria, as well as Candida albicans. Peptide 2 (P2) showed the most favorable activity profile and was selected for further characterization. Given the clinical relevance of Candida infections, subsequent characterization focused on the antifungal activity of P2. P2 was then tested against two C. albicans reference strains (SC5314 and ATCC 10231) and two non-albicans Candida clinical isolates (C. glabrata BG2 and C. parapsilosis Cp29). P2 inhibited both C. albicans strains and C. parapsilosis Cp29 at 62.5 µM, whereas C. glabrata BG2 was less susceptible (MIC, 250 µM). Hemolysis reached 12% at 500 µM, while no significant reduction in 293T-cell viability was observed up to 50 µM; marked cytotoxicity occurred at 500 µM. P2 retained antibacterial activity after exposure to 100°C for 15 min. Post-exposure viable-cell recovery showed strain-dependent effects, while flow cytometry revealed strain-dependent changes in CellROX Orange- and SYTOX Blue-positive populations, consistent with oxidative stress and compromised membrane integrity following P2 exposure.

P2 displayed antifungal activity against C. albicans and non-albicans Candida strains, with strain-dependent susceptibility and cellular responses. Its low hemolytic activity, limited cytotoxicity at lower concentrations, and retained antibacterial activity after heat exposure support further characterization of P2 as an antifungal peptide candidate.