Abstract
The ability to align pairs of very large molecular sequences is essential for a range of comparative genomic studies. However, given the complexity of genomic sequences, it has been difficult to devise a systematic method that can align - even within the same species - pairs of large sequences. Most existing approaches typically attempt to align nucleotide sequences while ignoring valuable features contained within them, eg they filter out low-complexity regions and retroelements before aligning the sequences. However, features are then added post-alignment for visualisation and analysis purposes. We argue that repetitive elements and other features (such as genes, exons and regulatory elements) should be part of the alignment process. A hierarchical approach that aligns the biologically relevant features before aligning the detailed nucleotide sequences has a number of interesting characteristics: (1) features define 'alignment anchor points' that can guide meaningful nucleotide alignment; (2) features can be weighted; (3) a hierarchical approach would identify only meaningful regions to be aligned; (4) nucleotide sequences can be described as sequences of features and non-features, providing a natural mechanism to divide the sequences for processing; and (5) computational speed is significantly faster than other approaches. In this paper, we describe and discuss a feature-based approach to aligning large genome sequences. We refer to this as 'feature-based sequence alignment'.