Nucleotide Sequence Similarity Search Using Techniques from Content-Based Image Retrieval

Abstract

The amount of DNA data continues to increase exponentially as a result of high- throughput next generation sequencing. Current state-of-the-art tools for nucleotide sequence similarity search are not equipped to deal with this growth and new thinking is needed to tackle the rising scalability challenges. This thesis investigates the experimental approach of translating DNA sequences into images and applying state of the art techniques from the field of content- based image retrieval to index and search the resulting images. The challenges of translating DNA sequences into images are discussed and two algorithms for image generation are proposed. We look into the different feature descriptors that are available and evaluate them in the context of the generated images. Lastly the approach as a whole is evaluated with the mean average precision metric using BLAST as the gold standard reference. The results show that the proposed approach is not successful in approaching BLAST in retrieval performance, but offers a significant reduce in index sizes and thus better performance and scalability on large DNA databases.