A Compression Model for DNA Multiple Sequence Alignment Blocks

A particularly voluminous dataset in molecular genomics, known as whole genome alignments, has gained considerable importance over the last years. In this paper, we propose a compression modeling approach for the multiple sequence alignment (MSA) blocks, which make up most of these datasets. Our method is based on a mixture of finite-context models. Contrarily to other recent approaches, it addresses both the DNA bases and gap symbols at once, better exploring the existing correlations. For comparison with previous methods, our algorithm was tested in the multiz28way dataset. On average, it attained 0.94 bits per symbol, approximately 7% better than the previous best, for a similar computational complexity. We also tested the model in the most recent dataset, multiz46way. In this dataset, that contains alignments of 46 different species, our compression model achieved an average of 0.72 bits per MSA block symbol.