Whole proteome mapping of compound-protein interactions

Area of Science:

• Drug Discovery and Development
• Computational Chemistry
• Bioinformatics

Background:

• Off-target drug binding frequently causes toxic side effects, leading to clinical trial failures.
• Experimental identification of unintended protein binders is costly due to the vast number of potential biological targets.
• Predicting compound-protein interactions is crucial for safe and effective drug development.

Purpose of the Study:

• To develop a computational strategy for proteome-wide mapping of compound-protein interactions.
• To enable high-throughput identification of drug targets and binding moieties.
• To create a user-friendly web server for drug-target identification.

Main Methods:

• A pipeline combining chemical similarity principles and deep learning was developed.
• Bioactive molecules were matched with chemically similar annotated "bait" compounds.
• Deep learning models were employed for ranking potential compound-protein interactions.

Main Results:

• A novel strategy for predicting off-target drug binding was successfully implemented.
• The developed pipeline enables proteome-wide mapping of compound-protein interactions.
• A web server, DRIFT, was constructed for efficient drug-target identification.

Conclusions:

• The proposed deep learning strategy offers a cost-effective alternative to experimental screening for identifying off-target effects.
• The DRIFT web server facilitates high-throughput, multi-ligand target identification.
• This approach aids in understanding drug toxicity and optimizing drug development pipelines.