Active Learning-Accelerated Discovery of Antifungal Polymers With Self-Assembly Capacity for Tackling Invasive Fungal Diseases.

Invasive fungal diseases (IFD) are life-threatening infections with limited clinical management. Antifungal peptide (AFP)-mimicking polymers are promising candidates for tackling IFD. However, due to the structural diversity of polymers, it remains a great challenge to discover new polymers with potent antifungal activity and biocompatibility. Herein, we propose a novel active learning-guided framework PolyCAML for accelerated discovery of antifungal polymers. Inspired by AFPs, we construct a quaternary copolymer library comprising 516,114 combinations with different side chains. By integrating automated synthesis platform, Graph Transformer machine learning model, and greedy algorithm, 11 top-performing candidates are screened via the Design-Build-Test-Learn (DBTL) intelligent iterative workflow within merely 18 days. Through further data mining, key factors that influence antifungal potency and biocompatibility of polymers are identified. By more rigorous biological evaluation, four target polymers are screened, exhibiting a minimum fungicidal concentration ≤ 8 µg/mL and a 50% inhibitory concentration of L929 cells ≥ 512 µg/mL. These polymers are self-assembled into positively charged nanoparticles for delivering fluconazole, which exhibit synergistic anti-biofilm activity by enhanced biofilm penetration. The in vivo therapeutic efficacy is confirmed by a fungemia model and a fungal keratitis model. Overall, we establish a multi-task intelligent prediction platform PolyCAML to discover antifungal polymers for tackling IFD.