3D Protein Structure Predicted from Sequence

Debora S. Marks, Lucy J. Colwell, Robert P. Sheridan, Thomas A. Hopf, Andrea Pagnani, Riccardo Zecchina, Chris SanderarXiv 2011 · arXiv

Abstract

Predicting the three-dimensional structure of a protein from its primary amino acid sequence remains a fundamental challenge in molecular biology. Marks et al. demonstrate a breakthrough method that leverages the evolutionary record to infer physical contacts. By applying a global statistical model to large multiple sequence alignments, the authors successfully disentangle direct residue couplings from indirect correlations. This approach allows for the de novo prediction of accurate 3D folds for diverse protein families, including challenging membrane proteins, using only sequence data.

Introduction

The functional capacity of a protein is inextricably linked to its three-dimensional architecture. While high-throughput sequencing has made genetic data abundant, experimental determination of protein structures via X-ray crystallography or NMR remains a bottleneck—slow, costly, and often technically unfeasible for membrane-bound proteins. This paper addresses the "inverse folding problem" by treating evolution as a massive, natural mutagenesis experiment.

The Challenge of Indirect Correlations

The central premise of evolutionary-based structure prediction is that residues in physical contact tend to co-evolve; a mutation in one residue often necessitates a compensatory mutation in its neighbor to maintain structural integrity. However, previous attempts to map these contacts were plagued by "noise" from indirect correlations. For instance, if residue A contacts B, and B contacts C, residues A and C will appear correlated even if they are far apart in 3D space. Distinguishing these direct physical contacts from transitive correlations is the key hurdle this research overcomes.

Methodology: Global Statistical Modeling

The authors' primary innovation is the transition from local to global statistical modeling.

  1. Direct Coupling Analysis (DCA): Instead of analyzing pairs of residues in isolation (e.g., via Mutual Information), the authors employed a maximum entropy model (a Potts model). This global approach considers the entire multiple sequence alignment (MSA) simultaneously, identifying the direct couplings that best explain the observed sequence variation across the whole family.
  2. Evolutionary Inferred Contacts (EICs): These high-scoring direct couplings are interpreted as physical distance constraints.
  3. Folding Pipeline: The EICs, combined with predicted secondary structures, are used as inputs for distance geometry and simulated annealing algorithms to generate candidate 3D folds.
  4. Blind Ranking: Candidate structures are ranked using an objective system based on physical principles, such as the characteristic handedness of alpha-helices, to identify the most likely native fold without reference to known structures.
Conceptual overview of inferring 3D contacts from correlated mutations in sequence alignments.

Key Insights

The success of this method rests on several critical observations:

  • Global vs. Local: A global model effectively "subtracts" indirect correlations, revealing the true physical architecture.
  • Information Threshold: Surprisingly, predicting every contact is unnecessary. The authors found that 0.5 to 0.75 predicted constraints per residue are sufficient to constrain the conformational search space and achieve a correct global fold.
  • Correcting Local Errors: The strength of the global evolutionary signal is often sufficient to override and correct errors in local secondary structure predictions.

Results and Validation

The method was validated on 15 diverse protein families from the PFAM database, encompassing small domains, large enzymes, and complex membrane proteins like Rhodopsin.

  • High Accuracy: For 80% of the tested families, the top-ranked predictions achieved a Cα-RMSD of 2.7–5.1Å over the majority of the protein.
  • Membrane Protein Success: The accurate prediction of Rhodopsin (4.8Å RMSD) is particularly significant, as membrane proteins are notoriously difficult to crystallize.
  • Data Scaling: The study highlights that the recent explosion in genomic data has finally provided the sequence depth (typically >1,000 divergent sequences) required for these statistical methods to succeed.
Predicted evolutionary inferred contacts (red lines) mapped onto a 3D protein fold.

Limitations and Future Directions

While a major step forward, the method has limitations. It is highly dependent on the availability of large, high-quality sequence alignments. Furthermore, while the global folds are correct, the current resolution is not yet at the atomic level required for high-precision drug design. Future work will likely focus on refining these models to achieve sub-2Å resolution and applying the method to the thousands of protein families whose structures remain unknown.

Summary updated

Abstract

The evolutionary trajectory of a protein through sequence space is constrained by function and three-dimensional (3D) structure. Residues in spatial proximity tend to co-evolve, yet attempts to invert the evolutionary record to identify these constraints and use them to computationally fold proteins have so far been unsuccessful. Here, we show that co-variation of residue pairs, observed in a large protein family, provides sufficient information to determine 3D protein structure. Using a data-constrained maximum entropy model of the multiple sequence alignment, we identify pairs of statistically coupled residue positions which are expected to be close in the protein fold, termed contacts inferred from evolutionary information (EICs). To assess the amount of information about the protein fold contained in these coupled pairs, we evaluate the accuracy of predicted 3D structures for proteins of 50-260 residues, from 15 diverse protein families, including a G-protein coupled receptor. These structure predictions are de novo, i.e., they do not use homology modeling or sequence-similar fragments from known structures. The resulting low Cα-RMSD error range of 2.7-5.1Å, over at least 75% of the protein, indicates the potential for predicting essentially correct 3D structures for the thousands of protein families that have no known structure, provided they include a sufficiently large number of divergent sample sequences. With the current enormous growth in sequence information based on new sequencing technology, this opens the door to a comprehensive survey of protein 3D structures, including many not currently accessible to the experimental methods of structural genomics. This advance has potential applications in many biological contexts, such as synthetic biology, identification of functional sites in proteins and interpretation of the functional impact of genetic variants.
BibTeX
@misc{marks2011d,
  author = {Debora S. Marks and Lucy J. Colwell and Robert P. Sheridan and Thomas A. Hopf and Andrea Pagnani and Riccardo Zecchina and Chris Sander},
  title = {{3D Protein Structure Predicted from Sequence}},
  year = {2011},
  eprint = {1110.5091},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1110.5091},
}
Figure