CERES: Cluster‑enabled regression of extract signatures for discovery of NRF2 activators in Centella asiatica by ion-mobility mass spectrometry, k-medoids clustering and ensemble Lasso regression.

Abstract

Discovery of bioactive phytochemicals from botanical extracts is a pivotal step for the modernization of traditional medicine and in the discovery of novel pharmaceuticals. However, traditional bioassay-guided fractionation is laborious and often leads to the re-identification of known compounds. To address these challenges, we have developed a novel discovery pipeline, called CERES (Cluster-Enabled Regression of Extract Signatures) that combines preparative column chromatography-based fractionation and bioactivity assessment, followed by loop-injection ion-mobility mass spectrometry, with machine learning. We apply cluster-based feature reduction and regularized regression models for rapid identification of bioactive phytochemicals. This approach reduces the need for analytical chromatographic separation in early dereplication stages, reducing bias and increasing speed, while remaining complementary to downstream structural confirmation workflows. By grouping mass spectrometry features into clusters representing different chemical entities derived from the same molecular species, the computational approach effectively reduces dimensionality and improves performance and interpretability of model results. The computational approach was successfully applied here to identify bioactive compounds in fractions of Centella asiatica water extract and uncovered bioactive mass-to-charge signals linked to known chemical features as well as new low-abundant, yet to be annotated signals to be investigated in further studies. This innovative approach offers a powerful and efficient way to advance the integration of traditional medicine with evidence-based practice by leveraging the power of ion-mobility mass spectrometry and machine learning.