Fast, accurate ranking of engineered proteins by target-binding propensity using structure modeling

Area of Science:

• Computational Biology
• Protein Engineering
• Biophysics

Background:

• Deep learning advances protein structure prediction accuracy.
• A gap exists between predicting protein structures and prioritizing binding potential for protein engineering.

Purpose of the Study:

• Introduce Automated Pairwise Peptide-Receptor Analysis for Screening Engineered proteins (APPRAISE).
• Develop a method to predict the target-binding propensity of engineered proteins.
• Bridge the gap between protein structure prediction and protein binder prioritization.

Main Methods:

• Utilize established structure prediction tools (e.g., AlphaFold-Multimer, ESMFold) to generate structural models.
• Perform rapid scoring analysis (under 1 CPU second per model) considering biophysical and geometrical constraints.
• Implement APPRAISE via a web-based notebook interface on Google Colaboratory.

Main Results:

• APPRAISE accurately classifies receptor-dependent vs. independent adeno-associated viral vectors.
• Successfully screened diverse engineered proteins, including miniproteins targeting SARS-CoV-2 spike, nanobodies, and peptides binding specific receptors (transferrin receptor, PD-L1).
• Demonstrated accuracy, interpretability, and generalizability of the APPRAISE method.

Conclusions:

• APPRAISE effectively predicts the target-binding propensity of engineered proteins.
• The method accelerates protein engineering for biomedical applications by enhancing the utility of protein structure prediction.
• APPRAISE offers a generalizable and interpretable solution for screening protein binders.