Exploring the potential of machine learning to design antidiabetic molecules: a comprehensive study with experimental...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Diabetes mellitus is a leading cause of mortality globally, necessitating novel therapeutic strategies.
• Data-driven approaches, powered by machine learning (ML), are revolutionizing the design of therapeutic agents.
• Developing effective inhibitors of alpha-amylase is a key strategy for managing diabetes.

Purpose of the Study:

• To employ machine learning-based small molecule design for identifying novel alpha-amylase inhibitors.
• To computationally predict and experimentally validate potential antidiabetic agents.
• To assess the inhibitory potential and thermodynamic stability of synthesized molecules.

Main Methods:

• Utilized chemoinformatics and binary fingerprint techniques to build ML models for alpha-amylase inhibitors.
• Applied ensemble-based ML predictions on molecular libraries and synthetic scaffolds.
• Synthesized top predicted molecules, including benzothiophene dioxolane derivatives, and evaluated their biological inhibitory properties.
• Performed thermodynamic simulations to analyze the stability and affinity of active compounds.

Main Results:

• Top 10 ML models achieved high performance metrics (avg. score 0.8216, Pearson-r 0.827, Q² 0.835).
• Synthesized molecules ALC5 and ALC6 exhibited potent alpha-amylase inhibition with IC50 values of 2.1 ± 0.14 µM and 5.71 ± 0.02 µM, respectively.
• Thermodynamic simulations confirmed the stability and affinity of the experimentally validated molecules.

Conclusions:

• Machine learning-driven design is a reliable and accurate method for discovering potential antidiabetic agents.
• The identified benzothiophene dioxolane derivatives show promise as therapeutic leads for diabetes management.
• Further exploration of ML in drug design is recommended for developing effective treatments.