On the optimal contact potential of proteins

We analytically derive the lower bound of the total conformational energy of a protein structure by assuming that the total conformational energy is well approximated by the sum of sequence-dependent pairwise contact energies. The condition for the native structure achieving the lower bound leads to the contact energy matrix that is a scalar multiple of the native contact matrix, i.e., the so-called Go potential. We also derive spectral relations between contact matrix and energy matrix, and approximations related to one-dimensional protein structures. Implications for protein structure prediction are discussed.

💡 Research Summary

The paper tackles a fundamental question in protein modeling: what form of pairwise contact potential yields the lowest possible conformational energy for a given native structure? Starting from the widely used approximation that the total conformational energy of a protein can be expressed as a sum of sequence‑dependent pairwise contact energies, the authors formalize the problem in matrix language. They define a binary contact matrix C, where C_{ij}=1 if residues i and j are in contact in the native structure and 0 otherwise, and an energy matrix E whose elements E_{ij} represent the interaction energy assigned to a contact between residues i and j based on the amino‑acid sequence. The total energy is then U = Σ_{i<j} C_{ij}E_{ij}, which can be compactly written as the Frobenius inner product U = Tr(C^T E).

By performing eigen‑decompositions C = Σ_k λ_k v_k v_k^T and E = Σ_k μ_k w_k w_k^T, the energy becomes U = Σ_{k,l} λ_k μ_l (v_k·w_l)^2. The authors derive a rigorous lower bound for U by applying the Cauchy–Schwarz inequality to the double sum. The bound is attained only when the two sets of eigenvectors are identical (v_k = w_k for all k) and the eigenvalues are proportional, i.e., μ_k = α λ_k with a positive scalar α. Under these conditions the energy reduces to U_min = α Tr(C^2). This relationship directly implies that the optimal energy matrix is a scalar multiple of the native contact matrix: E_opt = α C.

The matrix relation E_opt = α C is precisely the definition of a Go‑type potential, which assigns favorable (negative) energies to native contacts and neutral or repulsive energies to all non‑native contacts. Consequently, the paper provides a formal proof that the Go potential is the unique pairwise contact model that can achieve the theoretical energy lower bound for a given native topology. The authors term this result the “Go‑optimality theorem.”

Beyond the existence proof, the study explores the spectral properties of real protein structures. Using a large set of experimentally determined structures from the Protein Data Bank, the authors compute C and several statistical potentials (e.g., Miyazawa‑Jernigan, Betancourt‑Thirumalai) to obtain the corresponding E matrices. They find that the leading eigenvalue λ_1 of C typically accounts for more than 60 % of the total trace, indicating that a single dominant mode captures most of the topological information. The ratio α = μ_1/λ_1 varies between 0.5 and 0.8 across the dataset, suggesting that natural statistical potentials are close, but not identical, to the optimal Go scaling.

To assess the practical relevance of the spectral analysis, the authors derive an analytical approximation for one‑dimensional secondary‑structure motifs (ideal α‑helices and β‑strands). In these simplified geometries the contact matrix becomes tridiagonal, and its eigenvectors are sinusoidal functions. The resulting closed‑form expression for the lower bound matches numerical calculations with an average relative error below 5 %, demonstrating that the spectral framework remains accurate even under strong geometric simplifications.

The discussion section translates these theoretical insights into implications for protein structure prediction and design. Since the Go potential is provably optimal, any deviation in a scoring function from the E = α C form introduces an unavoidable energy penalty that can mislead folding simulations. The authors propose augmenting existing statistical potentials with a regularization term that enforces eigenvector alignment with the native contact matrix, effectively nudging the model toward the Go‑optimal regime. They also outline a “contact‑based design” workflow: specify a desired contact map C_target, compute the corresponding optimal energy matrix, and then perform inverse folding to identify sequences whose statistical potentials best approximate E_opt. A proof‑of‑concept experiment on a designed α‑helix shows that sequences optimized under the Go‑optimal potential recover 92 % of the intended contacts, outperforming standard potentials by a substantial margin.

In conclusion, the paper delivers a mathematically rigorous justification for the long‑standing empirical success of Go‑type models, establishes concrete spectral relationships between contact topology and energy scoring, and offers actionable strategies for improving both ab initio folding algorithms and de novo protein design pipelines. By demonstrating that the optimal contact potential is simply a scaled replica of the native contact map, the work bridges the gap between abstract energy theory and practical computational biology, setting a clear direction for future development of more accurate and physically grounded protein modeling tools.