Estimating effective DNA database size via compression ⋆

Abstract. Search for sequence similarity in large-scale databases of DNA and protein sequences is one of the essential problems in bioinformatics. To distinguish ran-dom matches from biologically relevant similarities, it is customary to compute statistical P-value of each discov-ered match. In this context, P-value is the probability that a similarity with a given score or higher would appear by chance in a comparison of a random query and a random database. Note that P-value is a function of the database size, since a high-scoring similarity is more likely to exist by chance in a larger database. Biological databases often contain redundant, identical, or very similar sequences. This fact is not taken into account in P-value estimation, resulting in pessimistic estimates. One way to address this problem is to use a lower effective database size instead of its real size. In this work, we pro-pose to estimate the effective size of a database by its com-pressed size. An appropriate compression algorithm will ef-fectively store only a single copy of each repeated string, resulting in a file whose size roughly corresponds to the amount of unique sequence in the database. We evaluate our approach on real and simulated databases.