The intrinsic dimension of protein sequence evolution

Elena Facco, Andrea Pagnani, Elena Tea Russo, Alessandro LaioPLOS Computational Biology 2019 · DOI · OpenReview

Introduction

Protein sequences are far from random; they are the result of billions of years of evolutionary pressure, finely tuned to fold into specific structures and execute precise biological functions. While it is well-understood that mutations at different sites are often correlated to maintain structural integrity, the global degrees of freedom available to a protein family remain poorly defined. This study investigates the "Intrinsic Dimension" (ID) of protein sequence evolution—the number of independent directions a sequence can change without losing its functional essence.

Methodology

The researchers utilized a combination of bioinformatics and manifold learning to estimate the ID of protein families. The core of their approach involved:

  1. Distance Metrics: To quantify the divergence between sequences, three metrics were compared: a Modified Hamming Distance, a BLOSUM distance (incorporating amino acid substitution likelihoods), and a Normalized Edit Distance.
  2. TWO-NN Estimator: This algorithm was selected for its robustness in estimating ID from discrete data. By analyzing the ratio of distances between a sequence and its two nearest neighbors, the TWO-NN estimator avoids the need for a coordinate-based embedding.
  3. Boundary Correction: Given that sequence distances are naturally bounded, the authors developed a specialized Root-Mean-Square Deviation (RMSD) minimization procedure to ensure that ID estimates were not skewed by the finite size of the sequence space.

The study analyzed 27 diverse protein families from the Pfam database and compared them against synthetic sequences generated by HMMER (site-specific models), Direct Coupling Analysis (DCA; pairwise correlation models), and ProteinEvolver (phylogeny-aware simulations).

Key Findings: The "Magic Number" of Evolution

The analysis revealed that natural protein sequence evolution is surprisingly low-dimensional. Across diverse families, the ID consistently fell between 6 and 12, regardless of the protein's actual length. This suggests a universal constraint on how proteins explore sequence space.

A critical insight emerged when comparing natural sequences to synthetic models. While DCA models are highly effective at predicting protein structures by capturing pairwise correlations, they failed to replicate the low dimensionality of natural evolution, producing sequences with an ID of approximately 37. This discrepancy indicates that pairwise structural constraints alone are insufficient to explain the simplicity of evolutionary paths.

The missing link was identified as phylogeny. When sequences were generated using models that account for shared ancestry (the "family tree" of evolution), the resulting ID dropped to approximately 4, aligning closely with natural observations. This suggests that the low-dimensional "shape" of protein families is a historical footprint of branching descent from a common ancestor.

Implications for Protein Design

The discovery that protein evolution is constrained to a small number of dimensions has significant implications for artificial intelligence and generative modeling in biology. The authors propose that Intrinsic Dimension should serve as a standard "stress test" for new models. If a generative model—such as a Variational Autoencoder or a Protein Language Model—produces sequences with a high ID, it may be "hallucinating" evolutionary trajectories that nature does not actually take.

Conclusion

The intrinsic dimension of 6 to 12 represents a fundamental property of protein evolution. By demonstrating that phylogeny, rather than just physical folding constraints, drives this low dimensionality, this research provides a new lens through which to view the history and future design of biological molecules. Future work will be needed to determine if this range is truly universal across the entire protein universe and to uncover the specific biological drivers behind these precise values.

Summary updated

Abstract

It is well known that, in order to preserve its structure and function, a protein cannot change its sequence at random, but only by mutations occurring preferentially at specific locations. We here investigate quantitatively the amount of variability that is allowed in protein sequence evolution, by computing the intrinsic dimension (ID) of the sequences belonging to a selection of protein families. The ID is a measure of the number of independent directions that evolution can take starting from a given sequence. We find that the ID is practically constant for sequences belonging to the same family, and moreover it is very similar in different families, with values ranging between 6 and 12. These values are significantly smaller than the raw number of amino acids, confirming the importance of correlations between mutations in different sites. However, we demonstrate that correlations are not sufficient to explain the small value of the ID we observe in protein families. Indeed, we show that the ID of a set of protein sequences generated by maximum entropy models, an approach in which correlations are accounted for, is typically significantly larger than the value observed in natural protein families. We further prove that a critical factor to reproduce the natural ID is to take into consideration the phylogeny of sequences.
BibTeX
@article{facco2019intrinsic,
  author = {Elena Facco and Andrea Pagnani and Elena Tea Russo and Alessandro Laio},
  title = {{The intrinsic dimension of protein sequence evolution}},
  journal = {PLOS Computational Biology},
  year = {2019},
  doi = {10.1371/journal.pcbi.1006767},
}
Figure