Relational database index choices for genome annotation data

Latest genomics techniques such as ChIP-chip, ChIP-seq, and RNA-seq have induced an exponential growth in the volume of genome annotations. The effective mining of these data is an indispensible step in systems biology. In this work we consider conventional B-tree, R-tree and no index options for annotation data in MySQL, Oracle, and PostgreSQL databases. We validate theoretical considerations of applicability of different indexes by computational experiments with gene, repeat masker, expressed sequence tag, and single nucleotide polymorphism annotations of Homo sapiens chromosome 22. The running times for distance and overlap queries suggest that, with the exception of PostgreSQL B-tree for overlap queries, R-trees are superior to B-trees for indexing annotations, and thus may be a good choice for genome annotation databases.