Machine learning prediction of multiple distinct high-affinity chemotypes for α-synuclein fibrils

Area of Science:

• Neuroscience
• Medicinal Chemistry
• Computational Biology

Background:

• Alpha-synuclein aggregates are key biomarkers for neurodegenerative diseases like Parkinson's.
• Positron emission tomography (PET) imaging requires specific ligands to detect these aggregates.
• Current ligand discovery methods can be time-consuming and expensive.

Purpose of the Study:

• To develop a machine learning (ML) model for efficient identification of novel PET imaging ligands.
• To discover high-affinity binding compounds targeting alpha-synuclein aggregates.
• To validate the generalization capability of the ML model in ligand discovery.

Main Methods:

• Developed an ML model trained on a small dataset (<300 binding measurements).
• Employed scaffold-guided curation to select a diverse prospective compound set (30 compounds) from a large chemical library (140 million members).
• Performed experimental validation of the selected compounds.

Main Results:

• Identified five high-affinity binding compounds specific for alpha-synuclein aggregates.
• The ML model demonstrated robust generalization capabilities in predicting novel binders.
• The scaffold-guided curation efficiently narrowed down the search space from millions to a manageable set.

Conclusions:

• Machine learning, combined with scaffold-guided curation, is a powerful and efficient strategy for discovering novel PET imaging ligands.
• The identified ligands show potential for advancing the diagnosis and monitoring of alpha-synucleinopathies.
• This approach significantly accelerates the drug discovery pipeline for neurodegenerative diseases.