Scaffold and SAR studies on c-MET inhibitors using machine learning approaches.

Area of Science:

• Medicinal Chemistry
• Drug Discovery
• Computational Chemistry

Background:

• Numerous c-mesenchymal-epithelial transition (c-MET) inhibitors show anticancer potential but face challenges in clinical trials due to poor efficacy and drug resistance.
• The scaffold-based chemical space of small-molecule c-MET inhibitors remains largely unanalyzed, hindering rational drug design.

Purpose of the Study:

• To construct and analyze the largest dataset of c-MET inhibitors to date.
• To identify key structural features, scaffolds, and fragments associated with c-MET inhibitory activity.
• To reveal structure-activity relationship (SAR) patterns for guiding future drug discovery efforts.

Main Methods:

• Compiled a dataset of 2,278 molecules with varying c-MET inhibitory activity (IC50).
• Utilized t-distributed stochastic neighbor embedding (t-SNE) for dimensionality reduction of chemical diversity.
• Employed clustering, chemical space networks (CSNs), activity cliff analysis, and decision tree modeling.

Main Results:

• No significant differences in drug-like properties were found between active and inactive molecules.
• Identified common scaffolds (M5, M7, M8) and dominant fragments (pyridazinones, triazoles, pyrazines).
• A decision tree model pinpointed essential structural features: >=3 aromatic heterocycles, >=5 aromatic nitrogen atoms, and >=8 nitrogen-oxygen atoms.

Conclusions:

• The study provides a comprehensive analysis of c-MET inhibitor chemical space and SAR.
• Identified "safe bets" and "dead ends" for c-MET inhibitor development.
• The findings offer valuable insights for screening new compounds and optimizing existing c-MET inhibitors.