A Predictive Tool for Electrophilic Aromatic Substitutions Using Machine Learning.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Drug development involves synthesizing numerous compounds to optimize potency and pharmacokinetics.
• Generating large libraries of analogues from identified scaffolds is time-consuming.
• Late-stage functionalization (LSF) offers a way to rapidly produce analogues by selectively modifying C-H bonds.

Purpose of the Study:

• To develop a machine learning model for predicting regioselective electrophilic aromatic substitution sites.
• To enhance the efficiency of analogue synthesis in drug discovery.
• To overcome challenges in directing electrophilic reactions on drug-like molecules.

Main Methods:

• A machine learning model was developed to predict reactive sites for electrophilic aromatic substitution.
• The model utilizes the compound's SMILES string as input.
• Six quantum mechanics descriptors are employed to identify reactive sites.

Main Results:

• The machine learning model achieved 93% accuracy on an internal validation set.
• The model correctly predicted reactive sites for 90% of molecules in an external validation set.
• This demonstrates the model's robust predictive capability for late-stage functionalization.

Conclusions:

• The developed machine learning model accurately predicts reactive sites for electrophilic aromatic substitution.
• This tool can significantly accelerate the synthesis of drug analogues.
• The model offers a powerful computational approach to aid late-stage functionalization strategies in drug discovery.