Residue burial encodes a protein's fold

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• Protein structure determination is complex, governed by high-dimensional energy landscapes.
• Accurate low-dimensional representations of these landscapes are sought for efficient prediction.
• Existing methods for protein fold prediction have limitations in efficiency and accuracy.

Purpose of the Study:

• To investigate if residue core identity can serve as an efficient low-dimensional representation of protein structure.
• To compare the predictive efficiency of residue core identity against other established representations.
• To re-frame protein folding prediction as a problem of predicting residue core identity.

Main Methods:

• Developed a binary encoding for residue core identity (buried or not buried).
• Tested the efficiency of this representation in predicting protein backbone conformation.
• Compared core identity predictions with Cα contact maps and FoldSeek's 3Di embeddings.
• Evaluated fold quality prediction using sequence information alone.

Main Results:

• Residue core identity predicts protein backbone conformation more efficiently than tested representations.
• Core identity is 4x more efficient than previous bit-per-residue estimates for native fold encoding.
• It is 2x more efficient than Cα contact maps and 1.5x more efficient than FoldSeek's 3Di embeddings.
• Predicting residue burial from sequence alone provides a more accurate fold quality estimate than predicting pairwise contacts.

Conclusions:

• Residue core identity offers a highly efficient low-dimensional representation for protein structure.
• This finding simplifies and reframes the protein folding problem.
• The study highlights the potential of core identity for accurate and efficient protein structure prediction.