Alignment Free Sequence Similarity Estimation using Local Binary Pattern on DNA Trajectory Images

Abstract

Graphical representation of DNA sequences is a popular alignment-free sequence comparison technique for predicting gene families, discovering evolutionary relations, finding similar regions among protein structures, and reconstructing phylogenetic trees. Here, we proposed an alignment-free sequence comparison method, where DNA sequences were represented as two-dimensional trajectories in images. These images contain information of multi-level variation embedded in the trajectories. A micro-texture encoding scheme named Local Binary Pattern (LBP) was applied to these images for generating alignment-free (AF) feature descriptors. Three distance measures were applied to these AF descriptors and using these distance scores, phylogenetic trees were reconstructed. Then, the results were compared with phylogenetic trees produced by binary image-based representation method. It was revealed that the sequence comparison capability of LBP based DNA trajectory image representation is more accurate than binary-based DNA image representation system. It was also found that cosine distance is the most appropriate measure for calculating distance among AF descriptors during phylogenetic tree reconstruction. In addition, the accuracy of our method was also compared with some state of the art alignment-free sequence comparison algorithms. Contribution—In this paper, Local Binary Pattern (LBP) encoding scheme was applied on DNA trajectory images to capture multi-level variation, and accordingly to generate an alignment-free descriptor for measuring the similarities among DNA sequences.