Fuzzy clustering of physicochemical and biochemical properties of amino Acids

Indrajit Saha, Ujjwal Maulik, Sanghamitra Bandyopadhyay, Dariusz PlewczyńskiAmino 2011 · DOI

Abstract

In this article, we categorize presently available experimental and theoretical knowledge of various physicochemical and biochemical features of amino acids, as collected in the AAindex database of known 544 amino acid (AA) indices. Previously reported 402 indices were categorized into six groups using hierarchical clustering technique and 142 were left unclustered. However, due to the increasing diversity of the database these indices are overlapping, therefore crisp clustering method may not provide optimal results. Moreover, in various large-scale bioinformatics analyses of whole proteomes, the proper selection of amino acid indices representing their biological significance is crucial for efficient and error-prone encoding of the short functional sequence motifs. In most cases, researchers perform exhaustive manual selection of the most informative indices. These two facts motivated us to analyse the widely used AA indices. The main goal of this article is twofold. First, we present a novel method of partitioning the bioinformatics data using consensus fuzzy clustering, where the recently proposed fuzzy clustering techniques are exploited. Second, we prepare three high quality subsets of all available indices. Superiority of the consensus fuzzy clustering method is demonstrated quantitatively, visually and statistically by comparing it with the previously proposed hierarchical clustered results.

Introduction

In bioinformatics, amino acids are often represented as vectors of numerical values known as amino acid indices, which capture properties such as hydrophobicity, size, or charge. The AAindex database serves as the primary repository for these indices, currently housing 544 distinct entries. However, the increasing volume of data has led to significant redundancy and overlap among these indices.

Previous efforts to categorize these properties relied on hierarchical clustering, a "crisp" method where each index is assigned to exactly one group. This approach failed to account for the nuanced, overlapping nature of biochemical properties and left 142 indices unclassified. Consequently, researchers often resort to manual, trial-and-error selection of indices for protein sequence analysis. This study addresses these limitations by introducing a fuzzy clustering framework that accommodates the inherent overlap in biological data.

Methodology

The proposed Consensus Fuzzy Clustering (CFC) framework utilizes several advanced techniques to achieve robust partitioning:

  1. Fuzzy c-Medoids (FCMdd): Unlike standard clustering that calculates a mathematical "mean," medoid-based clustering selects an actual representative index from the database as the cluster center. The "fuzzy" aspect allows indices to have partial membership in multiple clusters, reflecting their multi-faceted biological roles.
  2. Evolutionary Optimization: Genetic Algorithms (GA) and Differential Evolution (DE) are employed to optimize cluster centers and automatically determine the optimal number of clusters ($K$), avoiding local optima.
  3. Consensus Mechanism: A majority voting procedure integrates results from multiple fuzzy algorithms to ensure a stable and reliable final classification.
  4. Distance Metric: A Pearson correlation-based distance measures the similarity between indices across the 20 standard amino acids.
Overview of the Consensus Fuzzy Clustering (CFC) methodology showing the two-phase approach for known and unknown cluster numbers.

The study utilized the AAindex1 database (544 indices), comparing results against the 402-index benchmark set established by Tomii and Kanehisa (1996).

Key Results and Discussion

The CFC method demonstrated significant improvements over traditional hierarchical and individual fuzzy clustering techniques.

Quantitative Superiority

The CFC method achieved a Minkowski Score of 0.3026 (where lower values indicate better performance) and a Kappa Index of 0.9502. Statistical analysis confirms that the consensus approach consistently outperforms standalone fuzzy algorithms like DEFCMdd or GAFCMdd.

Comparison of Minkowski Scores for different fuzzy clustering techniques. Panel (b) highlights the superior performance (lowest score) of the proposed CFC method.

Biological Categorization

While previous literature identified six main categories, the CFC approach revealed that eight clusters provide a superior statistical and biological fit. These categories include:

  1. Electric properties
  2. Hydrophobicity
  3. Alpha and turn propensities
  4. Physicochemical properties
  5. Residue propensity
  6. Composition
  7. Beta propensity
  8. Intrinsic propensities

By successfully categorizing the 142 previously unclustered indices, the study provides a more comprehensive map of the amino acid property space.

High-Quality Subsets

To facilitate efficient protein encoding, the authors derived three "gold standard" subsets: HQI8, HQI24, and HQI40. These subsets represent the most informative and least redundant indices, serving as a streamlined resource for researchers.

Conclusion

The Consensus Fuzzy Clustering framework effectively addresses the challenges of redundancy and overlap in the AAindex database. By moving beyond crisp boundaries, the method provides a more accurate representation of biochemical properties. Future work may involve applying this framework to other complex bioinformatics datasets, such as gene expression or protein-protein interaction networks. The software and processed database are available at http://sysbio.icm.edu.pl/aaindex/.

Summary updated

Abstract

In this article, we categorize presently available experimental and theoretical knowledge of various physicochemical and biochemical features of amino acids, as collected in the AAindex database of known 544 amino acid (AA) indices. Previously reported 402 indices were categorized into six groups using hierarchical clustering technique and 142 were left unclustered. However, due to the increasing diversity of the database these indices are overlapping, therefore crisp clustering method may not provide optimal results. Moreover, in various large-scale bioinformatics analyses of whole proteomes, the proper selection of amino acid indices representing their biological significance is crucial for efficient and error-prone encoding of the short functional sequence motifs. In most cases, researchers perform exhaustive manual selection of the most informative indices. These two facts motivated us to analyse the widely used AA indices. The main goal of this article is twofold. First, we present a novel method of partitioning the bioinformatics data using consensus fuzzy clustering, where the recently proposed fuzzy clustering techniques are exploited. Second, we prepare three high quality subsets of all available indices. Superiority of the consensus fuzzy clustering method is demonstrated quantitatively, visually and statistically by comparing it with the previously proposed hierarchical clustered results. The processed AAindex1 database, supplementary material and the software are available at http://sysbio.icm.edu.pl/aaindex/ .
BibTeX
@inproceedings{saha2011fuzzy,
  author = {Indrajit Saha and Ujjwal Maulik and Sanghamitra Bandyopadhyay and Dariusz Plewczyński},
  title = {{Fuzzy clustering of physicochemical and biochemical properties of amino Acids}},
  booktitle = {Amino},
  year = {2011},
  doi = {10.1007/s00726-011-1106-9},
}
Figure