Evolving Models of Biological Sequence Similarity

For computer scientists the problem of biological data retrieval has become synonymous with homology-based retrieval of primary gene sequence data and their associated protein products. This perspective is accessible to computer scientists, as primary sequence data is modeled as strings and fundamental algorithmic tools can be applied. However, by sticking with this formative foundation, we computer scientists are failing to recognize new challenges and opportunities arising in this rapidly changing arena. Most notably, sequence data represents a polymer; a complex molecule is itself composed of a linear sequence of smaller molecules (monomers). The monomers interact with each other, both with their local neighbors and at long range, forming secondary and tertiary structure. As more sophisticated similarity models incorporate elements of these three-dimensional chemical environments, they have the potential to supplant the use of primary structure homology as the foremost method of data retrieval.

1. Introduction

Almost as if no one was looking the character of problem solving in bioinformatics has changed from established methods suited to a scarcity of data to the development of new methods suited to a wealth of diverse data. Originally, when data was scarce, the foremost biological problem associated with sequence data was identifying unknowns. Today the complete genomes of over 600 organisms have been sequenced [1]. Although existing methods can always be improved, they do an admirable job of identifying the location of genes and associating them with known, or at least putative, functional details [2]. The next step requires understanding how individual genes and their protein products function in a much more systematic way. Where once scientists asserted that 90% of the genome was “junk”, the importance of noncoding RNAs and repeat regions that make up the junk may actually transcend the importance of protein coding genes [3]–[5].