It is an old paper from 2009 and most readers are probably familiar with it. So, please excuse our ignorance.
The original BWT allows searching of a pattern in one direction, namely, from right to left (backward search). However, for ef?cient approximate pattern searching, we need a data structure that can support searching in both directions (forward and backward search) and allow us to switch from forward to backward search or vice versa in the course of aligning the pattern. This would enable us to start matching a pattern from the middle, then extend the search in either directions and switch directions in the alignment process. A seemingly trivial solution is to keep two BWTs, one for the original sequence and the other for the reversal. Then forward search can be done via a backward search on the reversal of the pattern based on the latter BWT.
However, this solution cannot allow interleaving the backward and forward search. And it deals with two separate searching space, demanding a lot more memory (see Section III for details). We develop a new indexing data structure, called bi-directional BWT, to solve this problem. Based on bi- directional BWT, we develop a new short read alignment tool 2BWT, which is faster than all existing tools.
Oddly, even though the above paper is old, not too many bioinformatics papers outside T.W. Lam’s group cite it. Jared Simpson and Timo Beller are exceptions. We came across it, while trying to understand Ruibang’s latest paper “SOAP3-dp: Fast, Accurate and Sensitive GPU-Based Short Read Aligner” published in PLOS One today. It will be covered in a future commentary.
For source code and other details, please check DNA Indexing and Alignment Tools at Hong Kong University.