Visualizing Amino Acid Substitutions in Physicochemical Space for Machine Learning

Single amino acid substitutions are associated with certain inherited genetic disorders, such as sickle cell disease and hemochromatosis. To help sift through the large genomic databases now available, a new method of visualizing the physical and chemical changes as 3D vectors is proposed. In a principal component analysis space, the most important axes can be interpreted as "size," "hydrophobicity," and "charge." It can be shown that substitutions accessible by single nucleotide changes are strongly correlated with directions in this space, based on the codon position and mutation type. This illustrates the granular control available under the canonical genetic code for potentially beneficial mutations conferring a heterozygote advantage. This work can be applied to machine learning algorithms to better understand the underlying mechanisms behind these hereditary conditions. Other potential applications include the development of new therapeutics via rational protein design, as well as an improved theoretical basis for understanding directed and natural protein evolution.