Using experimental results of protein design to guide biomolecular energy-function development

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Computational models of macromolecules are vital in biochemistry but limited by physical inaccuracies.
• Classical mechanics-based energy functions are commonly used but trained on limited datasets.
• Developing accurate energy functions is crucial for advancing protein design and modeling.

Purpose of the Study:

• To explore a novel paradigm for training computational energy functions using de novo protein design.
• To identify and address failure modes in the Rosetta energy function through experimental validation.
• To improve the accuracy of macromolecular modeling by refining energy function parameters.

Main Methods:

• Designed de novo proteins using the Rosetta energy function.
• Experimentally tested protein designs for stable folding and identified unstable predictions.
• Employed deep mutational scanning to pinpoint stabilizing mutations and understand energy function failure modes.
• Analyzed Rosetta's performance in refining protein crystal structures, focusing on core packing.

Main Results:

• Identified steric clashing and overpacking as key failure modes in the Rosetta energy function.
• Refined energy function parameters based on experimental data, specifically targeting native-like packing.
• Successfully reduced the identified failure mode in structure refinement while maintaining performance on other benchmarks.
• Developed an updated version of the Rosetta energy function with improved accuracy.

Conclusions:

• Experimental feedback from de novo protein design is a powerful tool for guiding energy function development.
• The refined Rosetta energy function demonstrates enhanced accuracy in predicting protein stability and structure.
• This work highlights the synergy between computational design, experimental validation, and energy function optimization.