Background: Bioactive natural products represent a vital resource for combating human diseases. However, their discovery often encounters multiple challenges. Bioactivity-guided isolation can yield bioactive compounds but are labor-intensive and have a low hit rate. In contrast, affinity-based ligand fishing enables rapid screening; however, the identified compounds are often structurally monotonous and exhibit weak bioactivity. Therefore, there is an urgent need for a novel strategy for rapid and targeted discovery of high-quality bioactive natural products.
Results: This study introduces a novel machine learning-assisted affinity ultrafiltration (ML-AAUF) strategy for the screening of bioactive natural products. Machine learning models were first trained and introduced to explore the chemical spaces. Using ML-AAUF, we identified nine compounds with neuraminidase inhibitory activity from Polygonum cuspidatum and Lonicera japonica. Flavonoids are common natural inhibitors of neuraminidase. In this study, three other structural types of neuraminidase inhibitors, including stilbene compounds, anthraquinone compounds and phenolic acid compounds containing ester bonds were also identified. Notably, resveratrol exhibited significant antiviral activity against H1N1 PR8, with an IC50 of 16.8 μM, highlighting its potential as an anti-influenza agent.
Significance: Machine learning-assisted affinity ultrafiltration strategy was first proposed. This strategy integrates machine learning's predictive capabilities with the rapidity and specificity of affinity ultrafiltration, offering a rapid and targeted approach to high-quality natural product discovery.
Keywords: High-quality natural products; LC-MS; Machine learning; Medicinal herbs; Neuraminidase inhibitors; Ultrafiltration.
Copyright © 2025. Published by Elsevier B.V.