Multipattern Consensus Regions in Multiple Aligned Protein Sequences and Their Segmentation

David K. Y. Chiu, Yan WangEURASIP J. Bioinform. Syst. Biol 2006 · DOI

TL;DR

This paper introduces "multipattern consensus regions" as a novel way to segment protein sequences by analyzing statistical variation patterns beyond simple conservation. By applying this method to the p53 tumor suppressor protein, they demonstrate that these regions correlate with known functional sites, 3D structural locations, and cancer mutation patterns, offering a more nuanced understanding of protein function and disease.

What Problem Does This Paper Address?

Understanding the functional regions within a biological sequence is crucial for deciphering its role. Traditional methods often rely on identifying conserved patterns (like the TATA box in DNA) or simple frequency counts. However, these approaches can miss more complex relationships within sequences, such as interdependent patterns where the presence of one amino acid influences the likelihood of another. The authors aim to develop a more comprehensive method for sequence segmentation that captures both conserved and interdependent statistical variations, providing a richer representation of functional regions. This is particularly relevant for proteins like p53, which are known to have complex functional roles and are frequently implicated in cancer.

Methods

The core of this paper is the introduction of "multipattern consensus regions" and a segmentation algorithm to identify them.

  • Multipattern Consensus Regions: These are defined as segments within multiple aligned sequences that are characterized by statistically significant conserved or interdependent patterns. This generalizes the concept of a simple consensus sequence by incorporating more complex statistical relationships.
  • Statistical Variation Patterns: The authors classify patterns at each aligned site into four categories:
  • Invariant (I): All outcomes are the same.
  • Conserved (C): Most outcomes are dominated by a single type, but not invariant.
  • Interdependent (D): Values are strongly associated with other values.
  • Hypervariant (V): Does not fit the above categories.
  • Measures:
    • Conserved Patterns: Measured using compositional redundancy (R(1)), based on Shannon entropy. Higher R(1) indicates greater conservation.
  • Interdependent Patterns: Measured using a function FD'(Xᵢ), which sums the statistically significant self-mutual information between a specific amino acid at position i and its associated amino acids at other positions.
  • Segmentation Algorithm: A heuristic algorithm is proposed, consisting of five phases:
  • Initiation: Regions are identified based on a majority pattern type within a sliding window.
  • Merging: Adjacent regions are merged if a statistical test of independence cannot distinguish them.
  • Region Type Determination: The type of each region is assigned based on the majority pattern within it.
  • Refinement: Boundaries and pattern types are refined, potentially merging regions of the same type if they are close and reapplying statistical tests.
  • Boundary Adjustment: Sites adjacent to boundaries that differ in type are removed.

Datasets

The primary dataset used for evaluation is the p53 protein sequence.

  • p53 Protein Sequences: Sequences from 31 species were retrieved from the SWISS-PROT database. These sequences were aligned using the ClustalW program.
  • Cancer Mutation Data: Data on p53 mutations and associated cancer types were obtained from the International Agency for Research on Cancer (IARC) database, containing 14,050 records.
  • Hereditary Factor Studies: The results were also compared to findings from twin studies on cancer etiology.

Key Insights and Ideas

  • Beyond Simple Conservation: The paper highlights that functional regions in proteins are not solely defined by conserved amino acids but also by complex interdependent relationships between amino acids.
  • Multipattern Consensus Regions as Functional Proxies: The identified multipattern consensus regions (I, C, D, V) appear to represent distinct functional or structural characteristics of the protein.
  • Structural Correlation: The study found that conserved (I) and interdependent (D) regions have distinct spatial localizations within the 3D structure of p53 – I-regions are buried internally, while D-regions are on the exterior. This suggests a link between statistical patterns and structural roles.
  • Cancer Linkage: The identified regions show significant associations with specific cancer types, suggesting that these patterns might be indicative of genetic predispositions or functional sites critical for cancer development. This provides a more localized view of where hereditary factors might manifest.

Key Results

  • The segmentation algorithm identified 18 distinct regions in the p53 core domain, comprising 7 D-regions, 5 I-regions, and 6 V-regions.
  • D-regions were predominantly found at the exterior of the p53 molecule, while I-regions and V-regions were buried within the interior.
  • A significant correspondence was found between mutated codons in I-regions and cancers of the stomach, colon, rectum, liver, and hematopoietic systems. D-regions were associated with cancers of the mouth, sinuses, and head/neck. V-regions were linked to testis and breast cancers.
  • The study's findings on region-cancer associations showed significant overlap with hereditary factors identified in human twin studies, reinforcing the biological relevance of the detected patterns.

Limitations and Future Work

  • Sample Size: The authors acknowledge that the sample size for the cancer association study was small, and further evaluation is needed to confirm these findings.
  • Generalizability: While the method is applied to p53, its applicability to other protein families and to DNA/RNA sequences requires further exploration.
  • Computational Complexity: The optimization of the segmentation algorithm can be computationally explosive, necessitating the use of heuristic approaches.
  • Future Work: The authors suggest that further evaluation of the detected patterns and their corresponding sites in relation to cancer associations is warranted. They also imply that this approach could be extended to other biosequence types and potentially lead to more refined functional annotations.

Summary updated

BibTeX
@article{chiu2006multipattern,
  author = {David K. Y. Chiu and Yan Wang},
  title = {{Multipattern Consensus Regions in Multiple Aligned Protein Sequences and Their Segmentation}},
  journal = {EURASIP J. Bioinform. Syst. Biol.},
  year = {2006},
  doi = {10.1155/bsb/2006/35809},
}
Figure