The Hunt for Genomic Dark Matter: Aligning Non-coding Functional DNA

Motivation: Statistical analysis of conservation patterns in human and mouse genomes suggests that as much as 5% of human genomic DNA is under purifying selection. Known genes account for 1.5%. Identifying and characterising the remaining 3.5% is still a major challenge. With the availability of more genomes, an intuitive way to detect conservation patterns is to align the human genome with genomes of other species such as mouse or rat. Compared to coding DNA, aligning non-coding DNA is a hard task because of the relatively high sequence divergence and the absence of codon structure allowing a higher incidence of insertions and deletions. So far, it has been difficult to assess or compare the performance of alignment algorithms because no suitable gold standards and no evaluation procedures have been proposed. Results: We propose an objective measure of the alignment accuracy of non-coding DNA called ”Gap Attraction”. ”Gap Attraction” gives a measure of the proportion of ungaps, the conserved regions between two random neighbouring indel events, that have been misaligned. The ”Gap Attraction” measure is derived from a model that assumes that insertions and deletions (indels) rain on the genome independently of each other and uniformly along the sequence. From that, it follows that the length of ungaps is geometrically distributed. This hypothesis is verified by the data for the alignment of human chromosome 21, and then the whole human genome, against the mouse genome. The histogram counts for ungaps of medium length lie within the 95% confidence intervals confirming the hypothesis. ”Gap Attraction” does not require knowledge of the true alignment. We measured the ”Gap Attraction” index for two widely-used alignment algorithms; Blastz which was developed specifically for aligning human-mouse DNA, and Clustalw; a global alignment algorithm for pairwise protein and DNA sequences. As expected, Blastz performs better than Clustalw according to our evaluation measure. Contacts: [email protected], [email protected]