Combining machine learning with structure-based protein design to predict and engineer post-translational...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics
• Protein Engineering

Background:

• Post-translational modifications (PTMs) significantly impact protein function, stability, and interactions, offering vast potential for protein engineering.
• Over 400 PTMs are known, presenting both challenges and opportunities for diversifying protein design beyond genetically encoded amino acids.

Purpose of the Study:

• To develop computational tools for predicting and designing specific protein post-translational modifications (PTMs).
• To integrate artificial neural network (ANN) predictions with the Rosetta protein modeling suite for structure-based PTM design.
• To enable the rational modification and introduction of novel PTMs in proteins for therapeutic applications.

Main Methods:

• Trained ANNs to predict eighteen common PTMs, including glycosylation, phosphorylation, methylation, and deamidation.
• Integrated ANN models into the Rosetta computational protein modeling suite.
• Developed a design protocol to control the probability of specific PTMs at particular sites.

Main Results:

• Successfully combined ANN predictions with Rosetta for structure-based PTM design.
• Demonstrated the ability to modify existing PTMs and introduce novel ones.
• Validated the potential for applications such as epitope masking and enhancing protein stability.

Conclusions:

• The developed computational approach provides novel tools for the rational design of PTMs within the Rosetta protein engineering framework.
• This methodology can significantly advance the design of protein therapeutics by precisely controlling PTMs to modulate their properties.
• The work expands the protein engineering toolbox, enabling tailored modifications for diverse biological applications.