Modeling Structure-Activity Relationship of AMPK Activation.

Area of Science:

• Biochemistry
• Computational Biology
• Pharmacology

Background:

• Adenosine monophosphate activated protein kinase (AMPK) regulates key cellular functions including metabolism and growth.
• AMPK activation is therapeutically beneficial in diseases such as metabolic syndrome, obesity, diabetes, and cancer.
• Natural compounds often activate AMPK indirectly, posing challenges due to complex mixtures and limited availability of pure substances.

Purpose of the Study:

• To develop a computational screening tool for identifying adenosine monophosphate activated protein kinase (AMPK) activators.
• To address the heterogeneity in AMPK activation pathways caused by complex natural compound mixtures.
• To enable rapid in silico screening of compounds and mixtures for potential therapeutic applications.

Main Methods:

• Machine learning algorithms including random forest, support vector machines, stochastic gradient boosting, logistic regression, and deep neural networks were applied.
• Models were trained and validated using a database of 904 activating and 799 neutral or inhibiting compounds.
• Data was sourced from extensive PubMed literature searches and the PubChem Bioassay database.

Main Results:

• All machine learning models demonstrated high classification accuracy during training.
• Crucially, the models exhibited strong predictive performance on unseen test data.
• The developed models are effective for rapid in silico screening of diverse chemical entities.

Conclusions:

• Machine learning provides a robust method for screening compounds that modulate adenosine monophosphate activated protein kinase (AMPK) activity.
• These computational tools can accelerate the identification of novel AMPK activators from complex mixtures.
• The validated models offer a valuable resource for discovering compounds for further experimental testing and therapeutic development.