Sequence-based drug design as a concept in computational drug design

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Traditional structure-based drug design (SBDD) is a complex, multi-step process.
• SBDD relies heavily on the availability of high-quality 3D protein structures.
• Developing drugs for challenging targets or proteins without known structures remains a significant hurdle.

Purpose of the Study:

• To propose and validate a novel sequence-to-drug concept for computational drug design.
• To develop a computational tool based on end-to-end differentiable learning for drug discovery.
• To demonstrate the utility of protein sequence information in identifying drug candidates and targets.

Main Methods:

• Developed TransformerCPI2.0, a deep learning model utilizing end-to-end differentiable learning.
• Trained and validated the model on protein sequence data for drug-target interaction prediction.
• Applied the model to discover novel drug hits for challenging targets and identified new targets for existing drugs.

Main Results:

• TransformerCPI2.0 demonstrated strong generalization capabilities across diverse proteins and compounds.
• The model successfully identified new drug candidates for difficult-to-target proteins.
• The inverse application of the concept led to the identification of a new target for an existing drug.

Conclusions:

• The sequence-to-drug concept provides a viable alternative to SBDD, particularly for proteins lacking 3D structural data.
• End-to-end differentiable learning from protein sequences offers a powerful new paradigm in computational drug design.
• This approach expands the scope of drug discovery by enabling the targeting of previously intractable proteins.