Pmf-cpi: assessing drug selectivity with a pretrained multi-functional model for compound-protein interactions

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Compound-protein interactions (CPI) are vital in drug development.
• Assessing drug selectivity is crucial for minimizing side effects.
• Existing selectivity models often lack sufficient data for specific targets.

Purpose of the Study:

• To introduce a pretrained multi-functional model for compound-protein interaction prediction (PMF-CPI).
• To fine-tune PMF-CPI for evaluating drug selectivity.
• To demonstrate PMF-CPI's capability in predicting drug-target interactions and selectivity.

Main Methods:

• Utilized recurrent neural networks and the TAPE language model for protein embedding.
• Employed a graph encoder to extract molecular information.
• Applied dense layers for output generation.
• Fine-tuned the model on datasets including human cytochrome P450s.

Main Results:

• PMF-CPI outperformed existing methods in binding affinity regression and CPI classification.
• The model accurately predicted varying drug affinities and interactions for similar targets.
• Demonstrated effective analysis of drug selectivity.

Conclusions:

• PMF-CPI offers a robust solution for predicting compound-protein interactions and drug selectivity.
• The model facilitates the identification of selective drugs for targeted therapies.
• PMF-CPI advances the development of safer and more effective therapeutics.