ProteinMPNN Recovers Complex Sequence Properties of Transmembrane β-barrels

Area of Science:

• Computational biology
• Protein engineering
• Biophysics

Background:

• Deep learning (DL) methods have advanced protein design, yet struggle with de novo membrane proteins (MPs) and complex beta-sheet folds.
• Current DL models require robust evaluation for challenging protein structures.

Approach:

• Benchmarked ProteinMPNN against Rosetta Franklin2023 for transmembrane and beta-barrel folds.
• Assessed the impact of input backbone refinement on ProteinMPNN's performance.
• Evaluated generated sequences using in silico filters for experimental validation.

Key Points:

• Refined backbone inputs significantly improve ProteinMPNN's accuracy in capturing global sequence properties for complex folds.
• ProteinMPNN generates more diverse transmembrane beta-barrel sequences than Franklin2023, particularly in pore-facing regions.
• ProteinMPNN-designed sequences show potential for experimental validation in nanopore applications.

Conclusions:

• ProteinMPNN's limitations in designing beta-sheet proteins are attributed to input backbone accuracy, not inherent software constraints.
• The study highlights the potential of DL methods for de novo design of functional membrane proteins and nanopores.