Exploring the potential of compound-protein complex structure-free models in virtual screening using BlendNet.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Identifying novel drug compounds is a bottleneck in early drug discovery.
• Structure-based affinity prediction models require complex 3D structures, limiting their use.
• Predicting binding affinity without 3D structural data is challenging.

Purpose of the Study:

• To introduce BlendNet, a novel framework for compound-protein affinity prediction.
• To improve affinity prediction accuracy by leveraging knowledge transfer.
• To overcome the limitations of 3D structure-dependent models.

Main Methods:

• Developed BlendNet, a framework utilizing a knowledge transfer strategy.
• Trained BlendNet to learn interdependent relationships between compounds and proteins.
• Evaluated BlendNet's performance against state-of-the-art models.

Main Results:

• BlendNet demonstrated superior performance in affinity prediction.
• Achieved improved accuracy across various cold-start scenarios.
• Showcased the ability to interpret interactions without 3D complex structures.

Conclusions:

• BlendNet offers a promising alternative to structure-based affinity prediction.
• The framework has the potential to accelerate and streamline drug development.
• BlendNet effectively predicts binding affinity without relying on 3D structural information.