Computational Methods for Predicting Chemical Reactivity of Covalent Compounds.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Covalent inhibitors offer therapeutic advantages like prolonged efficacy and precise targeting.
• However, their inherent reactivity can cause off-target effects and toxicity, necessitating accurate prediction.
• Modulating and predicting covalent compound reactivity is crucial for safe drug development.

Purpose of the Study:

• To compile a dataset of cysteine-targeted covalent compounds and their reactivity.
• To develop and validate computational models for predicting covalent compound reactivity.
• To provide an efficient tool for guiding covalent drug discovery.

Main Methods:

• Extensive literature review to compile a dataset of 419 cysteine-targeted covalent compounds.
• Application of machine learning, deep learning, and quantum mechanical calculations.
• Development of FP-Stack models for reactivity prediction.

Main Results:

• FP-Stack models achieved high predictive accuracy with Pearson and Spearman correlations of ~0.80 and ~0.75 on the test set.
• The models enable rapid and accurate reactivity predictions, reducing computational costs.
• Experimental validation on acrylamide compounds confirmed the model's predictive efficacy.

Conclusions:

• An efficient computational tool for predicting covalent compound reactivity has been established.
• This tool can significantly streamline drug discovery and development processes.
• The findings offer valuable insights for designing safer and more effective covalent drugs.