ProDualNet: dual-target protein sequence design method based on protein language model and structure model

Area of Science:

• Computational biology
• Protein engineering
• Bioinformatics

Background:

• Proteins interacting with multiple partners are key to biological regulation.
• Multi-target peptide drugs show therapeutic promise, necessitating multi-target protein design strategies.
• Current protein design methods primarily focus on single-receptor interactions, limiting dual-target design capabilities.

Purpose of the Study:

• To introduce ProDualNet, a novel structure-based computational method for designing protein sequences that bind to two distinct receptors.
• To address the challenge of limited experimental data for dual-target protein structures.

Main Methods:

• ProDualNet utilizes a heterogeneous graph network for pretraining.
• The model is fine-tuned using a combination of noise-augmented single-target data and real dual-target data.
• It integrates sequence and structure information from two target receptors.

Main Results:

• ProDualNet demonstrated superior performance in recovery and success rates compared to existing multi-state design methods across various test sets.
• In silico evaluations showed potential for designing dual-target proteins, including cases with allosteric binding and non-overlapping interfaces.
• The method effectively overcomes limitations posed by scarce dual-target protein structural data.

Conclusions:

• ProDualNet represents a significant advancement in computational protein design for multi-target applications.
• The developed method holds promise for accelerating the design of novel therapeutic proteins and biologics.
• Availability of code and data facilitates further research and application in protein engineering.