Proteome-Informed Machine Learning Studies of Cocaine Addiction

Area of Science:

• Computational Biology
• Pharmacology
• Neuroscience

Background:

• Decades of research have failed to yield FDA-approved anti-cocaine addiction drugs.
• Cocaine addiction involves complex protein interactions around the dopamine transporter, difficult to study traditionally.

Purpose of the Study:

• To develop a proteome-informed machine learning (ML) platform for discovering novel anti-cocaine addiction compounds.
• To identify potential drug targets and screen existing and candidate drugs for efficacy and safety.

Main Methods:

• Analyzed proteomic protein-protein interaction networks in cocaine dependence to identify 141 drug targets.
• Built 32 ML models to analyze over 60,000 drug candidates for cross-target effects, side effects, and repurposing potential.
• Predicted ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties for drug candidates.

Main Results:

• The ML platform identified 141 drug targets involved in cocaine dependence.
• Cross-target and ADMET screenings revealed that most existing drug candidates are unsuitable.
• Several nearly optimal lead compounds for anti-cocaine addiction were identified for further development.

Conclusions:

• The proteome-informed ML platform offers an innovative strategy for anti-cocaine addiction drug discovery.
• This approach effectively screens large numbers of compounds, overcoming limitations of traditional experimental methods.
• The study identified promising lead compounds, paving the way for future therapeutic development.