Automated Deep Learning-Based Pipelines for Multi-Objective De Novo Protein Design

Area of Science:

• Computational biology
• Protein engineering
• Deep learning in structural biology

Background:

• Deep learning models have revolutionized protein structure prediction and sequence design.
• Accurate prediction and design are crucial for engineering novel protein functions.

Purpose of the Study:

• To present a detailed protocol for EvoPro, an automated platform for protein design.
• To enable engineering of protein-protein interactions with customizable properties.

Main Methods:

• EvoPro utilizes a genetic algorithm combined with iterative structure prediction (AlphaFold2/AlphaFold3) and sequence design (ProteinMPNN/LigandMPNN).
• The protocol details multistate design objectives for simultaneous optimization of positive and negative design goals.
• Includes setup for genetic algorithm, scoring functions, and result analysis.

Main Results:

• The platform builds on a validated approach successfully generating high-affinity binding domains without experimental optimization.
• Demonstrates adaptability for diverse objectives like binding site targeting, conformational specificity, and symmetric assembly.

Conclusions:

• EvoPro provides a user-friendly framework for leveraging deep learning in complex protein design challenges.
• The complete computational protocol is executable within a week by new users, facilitating advanced protein engineering.