Solvent accessibility, residue charge and residue volume, the three ingredients of a robust amino acid substitution matrix

Hani Goodarzi, Ali Katanforoush, Noorossadat Torabi, Hamed S. NajafabadiJournal of Theoretical Biology 2006 · DOI

TL;DR

This paper introduces a new amino acid substitution matrix, the SCV matrix, derived from solvent accessibility, residue charge, and residue volume. This matrix demonstrates superior performance in assessing the error-buffering capacity of the genetic code and in discriminating protein families compared to existing matrices, suggesting it's a more robust and unbiased tool for biological analyses.

What Problem Does This Paper Address?

Understanding the evolution of the genetic code is a long-standing challenge. One prominent hypothesis suggests the code evolved to minimize the impact of mutations and errors. To test this, researchers often use "cost measure matrices" or amino acid substitution matrices to quantify the "deleterious effect" of substituting one amino acid for another. However, many existing matrices are criticized for being "contaminated" by the very genetic code they are meant to analyze, making it difficult to draw unbiased conclusions about the code's optimality. This paper aims to address this by developing a novel, unbiased amino acid substitution matrix that can more reliably assess the error-minimizing properties of the genetic code and its role in protein family discrimination.

Methods

The core of this paper is the development and evaluation of the Solvent Accessibility, Residue Charge, and Residue Volume (SCV) matrix.

  • SCV Matrix Construction: The SCV matrix is built upon a linear relationship between three amino acid properties:
  • Solvent Accessibility: The degree to which an amino acid residue is exposed to the solvent.
  • Residue Charge: The electrical charge of the amino acid side chain.
  • Residue Volume: The physical volume occupied by the amino acid residue. The equation used is $g(a, a') = |S(a) – S(a')| + \alpha \cdot |V(a) – V(a')| + \beta \cdot |C(a) – C(a')|$, where $S$, $V$, and $C$ represent solvent accessibility, volume, and charge, respectively, and $\alpha$ and $\beta$ are constants determined through optimization.

  • Random Code Generation: To assess the genetic code's optimality, the authors employed two methods for generating random genetic codes:

  • Degenerate Method: This method allows for random assignment of codons to amino acids while preserving the degeneracy of the generated codes.
  • Classic Method: This method divides the 64 codons into 21 sets (corresponding to the 20 amino acids and stop codons) and randomly assigns amino acids to these sets, keeping the first two bases of codons invariant.
  • Optimality Measurement: The "load minimization" capacity of the genetic code was quantified using a fitness function ($ofaa$) and comparing the z-scores of the canonical code against those generated from random codes. A lower (more negative) z-score indicates better error minimization.
  • Code-Reflecting Matrices: To check for bias, the SCV matrix was compared to "code-reflecting matrices" (Angle measurement and CRM) that are derived directly from the structure of the canonical genetic code, using regression analysis.
  • Protein Family Discrimination: The SCV matrix was transformed into a scoring matrix suitable for local sequence alignment algorithms. Its ability to discriminate between protein families was then tested using pairwise alignments on representative sequences from various families, comparing its performance to established matrices like PAM and BLOSUM.

Datasets

  • Amino Acid Indices: The study utilized 516 amino acid indices downloaded from the AAINDEX database (available online at www.genome.ad.jp/dbget/aaindex.html). These indices represent various physicochemical properties of amino acids.
  • Protein Families: The Pfam-A database (version 18) and Swiss-Prot (release 48.5) were used to extract protein sequences belonging to 7973 families. This dataset was used for evaluating the SCV matrix's performance in protein family discrimination.

Key Insights and Ideas

  • Unbiased Nature of SCV: The SCV matrix, by combining fundamental physicochemical properties (solvent accessibility, charge, volume), appears to be unbiased towards the genetic code, unlike matrices derived from observed protein sequences or evolutionary data. This is a crucial step towards reliable analysis of genetic code evolution.
  • Solvent Accessibility as a Key Factor: The paper highlights solvent accessibility as a critical property that strongly influences protein structure and function, and thus mutation impact. Its inclusion in the SCV matrix is a key innovation.
  • SCV's Superior Error Minimization: The SCV matrix consistently revealed a higher degree of error minimization in the canonical genetic code compared to many other amino acid indices, suggesting it captures more relevant information about the code's design.
  • Robustness in Protein Family Discrimination: The SCV-derived scoring matrix demonstrated strong performance in distinguishing protein families, even those considered "hard-to-discriminate," rivaling or exceeding established matrices like PAM50. This indicates its practical utility beyond theoretical genetic code analysis.

Key Results

  • The SCV matrix was constructed with optimized parameters $\alpha = 0.5$ and $\beta = 0.188$, yielding a z-score of -2.130 for the canonical genetic code's error-buffering capacity. This z-score was found to be significantly lower (more efficient) than that obtained from 516 other amino acid indices.
  • Regression analysis showed low correlation coefficients between the SCV matrix and code-reflecting matrices (Angle measurement and CRM), supporting its lack of bias towards the genetic code.
  • In protein family discrimination tasks, the SCV-derived scoring matrix performed comparably to or better than established matrices like PAM50 and BLOSUM65, particularly in identifying "hard-to-discriminate" families.
  • The SCV matrix is symmetrical, meaning the cost of substituting amino acid A for B is the same as B for A, and all substitution costs are positive.

Limitations and Future Work

  • Parameter Optimization: While $\alpha$ and $\beta$ were optimized using the degenerate method, the paper acknowledges that other methods of random code generation might yield slightly different optimal values.
  • Simplification of Properties: The SCV matrix is a linear combination of three properties. The authors suggest that a more complex model might capture even finer differences between amino acids, potentially leading to more significant findings regarding the genetic code's optimality.
  • Future Work: The authors propose that the SCV matrix can be further refined and applied to a wider range of biological problems, including protein structure prediction and evolutionary analysis, to leverage its unbiased nature and robust performance. The paper also hints at the potential for developing more sophisticated matrices by incorporating additional physicochemical properties.

Summary updated

BibTeX
@article{goodarzi2006solvent,
  author = {Hani Goodarzi and Ali Katanforoush and Noorossadat Torabi and Hamed S. Najafabadi},
  title = {{Solvent accessibility, residue charge and residue volume, the three ingredients of a robust amino acid substitution matrix}},
  journal = {Journal of Theoretical Biology},
  year = {2006},
  doi = {10.1016/j.jtbi.2006.12.014},
}
Figure