Machine Learning-Based Bioactivity Prediction and Descriptor-Guided Rational Design of Amyloid-β Aggregation...

Area of Science:

• Computational chemistry
• Drug discovery
• Neuroscience

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by amyloid-beta (Aβ) aggregation.
• Inhibiting Aβ aggregation is a key therapeutic strategy, but its complex nature complicates inhibitor design.
• Current experimental methods for developing Aβ inhibitors are slow and resource-intensive.

Purpose of the Study:

• To develop a machine learning-based tool for rapid virtual screening of small molecules targeting Aβ aggregation.
• To overcome the limitations of conventional and computational inhibitor design for Aβ.
• To accelerate the discovery of novel Alzheimer's disease therapeutics.

Main Methods:

• Developed Amylo-IC50Pred, a user-friendly web platform integrating machine learning models.
• Trained two classification models and one regression model on 584 biologically validated compounds.
• Utilized Random Forest and Histogram-based Gradient Boosting algorithms for prediction.

Main Results:

• The Random Forest model achieved 100% accuracy in distinguishing inhibitors from decoys.
• Histogram-based Gradient Boosting accurately classified inhibitor potency with 81% accuracy.
• The Random Forest regression model achieved a high R² of 0.93 for predicting IC50 values.

Conclusions:

• Amylo-IC50Pred provides a rapid and accurate method for virtual screening of Aβ aggregation inhibitors.
• Key molecular properties like hydrophobicity and shape are crucial for effective Aβ inhibition.
• The platform serves as a valuable resource for accelerating Alzheimer's disease drug discovery.