Abstract
Phenolic compounds (PC) are important antioxidant biomolecules for medicine and foods industries. The DDPH test is used for testing antioxidant capacity. A fully No-Code methodology is presented for building QSPR models for the anti-DPPH activity of 202 PC. Machine learning (ML) algorithms were used for dimensionality reduction (PCA, InfoGain, GainRatio, CfsSubset) and predictive model training (J48, RandomTree, JCHAID*). Conceptual Density Functional Theory (CDFT) descriptors are calculated at the GFN1-xTB and GFN2-xTB levels of theory and the resulting global reactivity descriptors are used to train the ML models. The obtained Decision Tree (DT) models all present over 85% accuracy and Substantial Agreement with Reality, both for the internal and external validation. All the developed models adhere to the OECD guidelines for regulatory QSPR developments and are discussed in a mechanistic context. This research presents a novel, simple and codeless methodology for developing highly precise predictive models for the anti-DPPH activity of PC, successfully bridging the gap between experimental chemistry, theoretical physical chemistry, and ML.
Original language | English |
---|---|
Article number | 114782 |
Journal | Computational and Theoretical Chemistry |
Volume | 1239 |
DOIs | |
Publication status | Published - Sept 2024 |
Keywords
- Antioxidant mechanism
- Artificial intelligence
- CDFT
- DPPH
- ML
- Phenolic compounds
- XAI
ASJC Scopus subject areas
- Biochemistry
- Condensed Matter Physics
- Physical and Theoretical Chemistry