TL;DR
This paper introduces a novel method for characterizing amino acid environments by analyzing the radial distribution of their surrounding physical and chemical features. This approach leads to a new amino acid similarity matrix (WAC) that shows promise for improving protein sequence alignment sensitivity, particularly for longer sequences.
What Problem Does This Paper Address?
Understanding how amino acids interact within a protein's microenvironment is crucial for predicting protein structure, function, and the impact of mutations. Traditional methods for comparing amino acids often rely on empirical substitution frequencies (like BLOSUM and PAM matrices) or direct comparisons of physico-chemical properties. However, these approaches may not fully capture the nuanced spatial relationships of features around an amino acid, which can significantly influence its compatibility within a protein. This paper aims to address this by developing a more detailed representation of amino acid environments that explicitly considers the geometry and distance of surrounding features.
Methods
The core of this work is the FEATURE system, which analyzes protein microenvironments. For each amino acid, the system defines a "site" centered on its Cβ atom. This site is then divided into concentric radial shells, each 1 Å thick, extending up to 10 Å. Within these shells, 21 different features are analyzed, encompassing atomic composition, chemical groups, residue properties, and secondary structure.
The key innovation is the radial distribution of these features. Instead of just knowing a feature exists, the system quantifies where it is relative to the amino acid's Cβ atom. This allows for a more detailed comparison of environments. To assess the significance of these features, a non-parametric Wilcoxon Rank Sum test is used to compare the observed feature distributions in amino acid sites against those in random "nonsites."
Based on these detailed environmental comparisons, the authors construct a new amino acid similarity matrix called the WAC matrix. This matrix is derived from a statistical summary of the squared differences in feature distributions between pairs of amino acids across all radial shells.
Datasets
The amino acid data for this study was extracted from a set of 20 proteins from the Protein Data Bank (PDB). These proteins were selected based on specific criteria: resolution finer than 3Å, no significant structural similarities, and no homologous sequence or function. The average sequence length of these proteins was 268 amino acids.
For evaluating the WAC matrix, 39 protein families were randomly selected from the PROSITE database. For each family, the most distant protein sequence was chosen as a query for BLAST searches.
Key Insights and Ideas
- Geometry Matters: The spatial arrangement and distance of features around an amino acid are critical for defining its environment and its compatibility with other residues. Simply knowing a feature exists isn't enough; knowing where it is relative to the amino acid's core is key.
- Radial Shells Reveal Nuances: Dividing the environment into radial shells allows for the detection of subtle differences between amino acids, even those that are closely related (e.g., Asp and Glu). This is because different amino acids might extend their influential features to different distances.
- WAC Matrix: A New Perspective: The WAC matrix, derived from these radial environmental comparisons, offers a different basis for sequence alignment than traditional matrices. It reflects the environmental compatibility of amino acids rather than just their evolutionary substitution rates or intrinsic physico-chemical properties.
- Potential for Improved Sensitivity: The WAC matrix demonstrates competitive performance with established matrices like BLOSUM62 and PAM250, and in some cases, shows improved sensitivity, particularly for longer protein sequences.
Key Results
- The WAC matrix was generated by comparing the radial distributions of 21 features for each amino acid against random controls and other amino acids.
- The WAC matrix showed a correlation of 0.8136 with BLOSUM62 and 0.7683 with Rao's matrix.
- In BLAST searches against 39 PROSITE families, WAC correctly identified more sequences than BLOSUM62 in 3 groups and more than PAM250 in 11 groups.
- For protein families with an average sequence length greater than 450 amino acids, WAC performed as well as or better than BLOSUM62.
- The authors observed that the WAC matrix appears to be more tolerant to mutations than BLOSUM62, as indicated by the diagonal elements of the difference matrix.
Limitations and Future Work
- Statistical Significance Threshold: The authors used a threshold of p=0.005 for statistical significance. They suggest that a stricter threshold could be applied to identify more defining features for amino acid comparisons.
- Rotational Freedom: For some amino acids with flexible side chains, the analysis averages over all possible conformations. The authors acknowledge that for specific protein contexts, modeling a subset of conformations might be more appropriate.
- Sequence Length Dependence: While WAC performed well on longer sequences, it did not perform as well on shorter sequences compared to BLOSUM62. This suggests that the environmental information captured by WAC might be more impactful in larger structural contexts.
- Future Directions: The authors propose that their environmental attributes could be used for fold recognition and for creating scoring functions in protein threading techniques. They also suggest that context-dependent similarity matrices could be generated by using specific sets of proteins or regions within proteins. Further refinement of the WAC matrix and its application to a broader range of biological problems are also envisioned.
Summary updated