Dimension Reduction of Amino Acid Indices with Principal Component Analysis

Amino acids are the building blocks of proteins (Figure 1). Physicochemical and biochemical attributes of amino acids play a vital role in protein structures and functions. Each amino acid index is an array of 20 numerical values demonstrating different properties of amino acids. Kawashima et al. [1] collected 544 amino acid indices from literature and published in AAindex database, which has risen to 566 indices. Tommi et al. [2] classified amino acid indices into six groups based on alpha and turn propensities, betapropensity, composition, hydrophobicity, physicochemical properties, and other properties. Yet large scale of the indices space confines computational research to limited number of indices. Dimension reduction methods reduce the size of data by extracting relevant information and disposing rest of data as noise. On average there is an absolute value for the correlation coefficient of 0.8 or larger between each index and at least 11 other indices. This suggests that there is a potential for dimensionality reduction in amino acid indices space.