Compressing Similar Biological Sequences Using FM-Index

Nowadays, decreasing cost and better accessibility of sequencing methods have enabled studies of genetic variation between individuals of the same species and also between two related species. This has led to a rapid increase in biological data consisting of sequences that are very similar to each other, these sequences usually being stored together in one database. We propose a compression method based on Wavelet Tree FM-index optimized for compression of a set of similar biological sequences. The compression method is based on tracking single changes (together with their context) between every single sequence and the chosen reference sequence. We call our compression method BIO-FMI. The space complexity of our self-index is O(n+n log σ+N+N log σ+N' (logr+logN'/r) + N'+N' log n+r log N'+r log N) bits when applied on a set of r sequences, where n is the length of the reference sequence, N is the total length of distinct segments in all sequences, N' is the count of distinct segments in all sequences and σ is the size of the alphabet. BIO-FMI distinguishes so-called primary occurrences (occurring in the reference sequence) and secondary occurrences (not occurring in the reference sequence). BIO-FMI can locate each primary occurrence in O(s log σ + r log N'/r) time and each secondary occurrence in O(s log σ), where s is the length of a sample with a localization pointer. BIO-FMI gives very promising results in compression ratio and in locate time when performed on an extremely repetitive data set (less than 0.5% mutations) and when the searched patterns are of smaller lengths (less than 20 bases). BIO-FMI is competitive in extraction speed and it seems to be superior in time needed to build the index, especially in the case when the alignments of single sequences are given in advance.