Benchmarking Inverse Statistical Approaches for Protein Structure and Design with Exactly Solvable Models

Area of Science:

• Computational biology
• Protein bioinformatics
• Statistical mechanics

Background:

• Inverse statistical approaches are increasingly used to infer protein structure and function from Multiple Sequence Alignments (MSA).
• The validity of the connection between inferred Potts Hamiltonians and actual protein energetics and structure has not been rigorously tested.
• Lattice protein models offer a controlled environment for benchmarking these computational methods.

Purpose of the Study:

• To benchmark inverse statistical approaches for protein modeling using a lattice protein model (LP).
• To investigate the relationship between inferred Potts Hamiltonians and the underlying protein structure and energetics.
• To assess the capability of inferred Potts models in designing novel protein sequences.

Main Methods:

• Constructing MSAs for sequences within specific LP structures.
• Inferring effective pairwise Potts Hamiltonians from the generated MSAs.
• Analyzing the inferred Hamiltonians to understand their dependence on native and competing protein folds.
• Utilizing inferred Potts models as Hamiltonians for de novo protein sequence design.

Main Results:

• Inferred Potts Hamiltonians successfully replicate key features of LP structures and energetics.
• Effective pairwise couplings reflect both stabilization of native folds (positive design) and destabilization of alternative folds (negative design).
• Potts models facilitate the generation of novel sequences with desired folds, outperforming independent-site models.

Conclusions:

• Inverse statistical approaches are effective for protein structure and function prediction from sequence data.
• The success of these methods is attributed to their ability to capture complex folding landscapes, including positive and negative design principles.
• Inferred Potts models provide a powerful framework for both understanding protein biophysics and designing new proteins.