PED: a novel predictor-encoder-decoder model for Alzheimer drug molecular generation.

Area of Science:

• Artificial Intelligence in Drug Discovery
• Neuroscience and Pharmacology
• Computational Chemistry

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder with limited treatment options.
• Acetylcholinesterase (AChE) inhibitors are used for AD, but often cause side effects.
• Current AI models for molecular design can be inefficient, requiring additional optimization steps.

Purpose of the Study:

• To introduce PED, a cognitive-conditional molecular design model using a variational auto-encoder (VAE).
• To generate a molecular library for identifying potential AChE inhibitors for AD treatment with minimal adverse effects.
• To develop an efficient AI system for simultaneous property prediction and molecule generation.

Main Methods:

• Fine-tuning a pre-trained VAE model (PED) on AChE active compounds from Binding DB.
• Leveraging conditional molecular generation for targeted property-based molecule design.
• Utilizing molecular docking to verify the binding affinity of generated molecules to AChE.

Main Results:

• PED efficiently generates molecules with desired properties, outperforming benchmark methods.
• The model demonstrates effective learning of chemical space representations.
• Generated molecules showed strong binding affinity to AChE, indicating potential therapeutic efficacy.

Conclusions:

• The PED model offers an efficient and effective approach for designing novel AChE inhibitors for Alzheimer's disease.
• This AI-driven system accelerates the discovery of AD therapeutics with improved safety profiles.
• The study highlights the potential of advanced AI in enhancing drug discovery and cognitive capabilities.