Fast and Ultra-Capable Protein Design: Advancing the Frontier Through Atomistic SE(3)-Equivariance with Genie 3.

Area of Science:

• Computational biology
• Protein engineering
• Molecular modeling

Background:

• Current protein design methods struggle with high-affinity binders, multi-motif scaffolding, and stabilizing large proteins.
• Advancements focus on improving active site generation and aligning design with computational selection (in silico oracle).
• Recent shifts include all-atom models and replacing generative models with hallucination approaches, impacting efficiency and success rates.

Purpose of the Study:

• To bridge the gap between generative and hallucination approaches in protein design.
• To develop a novel diffusion model that enhances SE(3)-equivariance for protein structure prediction.
• To achieve state-of-the-art performance in various protein design tasks, including binder design and motif scaffolding.

Main Methods:

• Revisiting SE(3)-equivariance using a branched polymer model for protein structures.
• Developing Genie 3, a diffusion model incorporating these principles.
• Evaluating Genie 3 on binder design, motif scaffolding, and unconditional generation benchmarks.

Main Results:

• Genie 3 achieves state-of-the-art performance across multiple protein design tasks.
• The model demonstrates significantly improved computational efficiency compared to existing methods.
• Successfully designed a nanomolar binder for Nipah Glycoprotein G with a 12.5% success rate.

Conclusions:

• Genie 3 represents a significant advancement in protein design capabilities.
• The study highlights the renewed importance of SE(3)-equivariance in molecular modeling.
• The findings open new frontiers in designing complex and functional protein structures.