Machine Learning Prediction of Amino Acid Patterns in Protein N-myristoylation

Abstract

Abstract. Protein N-myristoylation is the lipid modification in which the 14-carbon saturated fatty acid binds covalently to N-terminal of virus-based and eukaryotic protein. In this study, we suggest an approach to predict the pattern of N-myristoylation signal using the machine learning system BONSAI. BONSAI finds rules in combination of an alphabet indexings and decision trees. Computational experiments with BONSAI classified amino acid residues depending on effect for N-myristoylation and found rules of the alphabet indexing. In addition, BONSAI suggested new requirements for the position of an amino acid in N-myristoylation signal.

Performance Improvement in Protein N-Myristoyl Classification by Bonsai With Insignificant Indexing Symbol

“…In the previous paper [6], we modified BONSAI in order that it enables the assigning of positions of amino acids from the N-terminus. When sequences that occupy the low selective positions for amino acid were given, the modified BONSAI in [6] have produced decision trees with large depths, similar to a decision list, such as the tree in Figure 3, in which all conditions for classifying positive and negative examples are reflected as node labels.…”

“…When sequences that occupy the low selective positions for amino acid were given, the modified BONSAI in [6] have produced decision trees with large depths, similar to a decision list, such as the tree in Figure 3, in which all conditions for classifying positive and negative examples are reflected as node labels. We reported 2 types of new rules in the previous paper [6]; however, it is difficult to interpret rules for N-myristoylated sequences with a decision tree having such a large depth. Futher, taking into account the fact that N-myristoylated sequences have the low selective positions for amino acid, we have further modified BONSAI by introducing a new concept called an "insignificant indexing symbol."…”

“…By analyzing binary patterns shown in the table in the Figure 3, we found a rule that classifies the given positive and negative examples [6]. However, the accuracy of this rule is not high-61.1% for positive examples and 92.0% for negative examples.…”

“…The results of the computational experiments show that this introduction improves the accuracy of decision trees particularly for positive examples, i.e., for sequences known to be Nmyristoylated. In addition, this introduction allows BONSAI to generate decision trees that have smaller depth and fewer numbers of nodes than the decision trees produced by the original BONSAI [6]. We further report the results of a comparison between the proposed method and an existing public method, demonstrating that our method performs better than the existing method with respect to the accuracy of extraction of both positive and negative examples despite shorter extraction time.…”

“…By using BONSAI, we carried out the computational experiment to establish new rules characterizing the difference between positive examples and negative examples of N-myristoylation sequences, and established the following 2 types of new rules: one, a rule that supports the existing N-myristoylation rule; the other, a rule that has not been discovered thus far [6]. Thus, the usefulness of BONSAI has been proved for the characterization of N-myristoylation sequences.…”