PRosettaC outperforms AlphaFold3 for modeling PROTAC ternary complexes.

Area of Science:

• Biochemistry and structural biology
• Computational chemistry and drug discovery

Background:

• Targeted protein degradation using Proteolysis-Targeting Chimeras (PROTACs) is a promising therapeutic strategy.
• Accurate prediction of ternary complex structures is crucial for rational PROTAC design but remains challenging.

Purpose of the Study:

• To systematically benchmark AlphaFold-3 and PRosettaC for predicting ternary complex structures.
• To evaluate the performance of structure prediction tools using both static and dynamic evaluation methods.

Main Methods:

• Benchmarking AlphaFold-3 and PRosettaC against 36 crystallographically resolved ternary complexes.
• Utilizing DockQ score for quantitative assessment of interface accuracy.
• Employing molecular dynamics simulations for dynamic evaluation of predicted models.

Main Results:

• AlphaFold-3 performance can be overestimated due to non-specific protein interactions.
• PRosettaC shows better geometric accuracy in select cases but struggles with linker sampling.
• Dynamic evaluation revealed transient high-accuracy conformations for PRosettaC models missed by static analysis.

Conclusions:

• Static benchmarking may overlook the importance of protein flexibility in ternary complex formation.
• Dynamic evaluation strategies are essential for a more accurate assessment of structure prediction tools.
• This study provides a refined framework for evaluating in silico tools for PROTAC development.