A major challenge posed by the rapid accumulation of whole-genome sequence data has been finding ways to interpret such data in a biologically meaningful way (i.e. finding an effective data mining technique that produces new biological insight). One of the first steps in DNA data mining is genome annotation: the process by which putative gene sequences are identified by establishing homology of open reading frames to existing genes. Current methods of homology searching can only partially provide annotation for most newly-sequenced genomes. Algorithms such as BLAST have inherent shortcomings which prevent them from detecting weak homology, especially when dealing with genomes of unusual composition. For my research project we are developing algorithms to improve such methods of homology searching.

We believe that Mutation biases (when the four types of DNA nucleotide do not mutate to one another with equal frequency) interact with the biochemical properties of amino acids and the codon assignments of the genetic code to produce complex variation in the patterns by which amino acids substitute for one another in different genes and different genomes.

We plan to use this knowledge to create biochemical profiles that would allow us to predict the way a gene would “look” in a different genome.

Next, we plan to extend the sort of search carried out by PHI-BLAST (a modified version of BLAST) where instead of a query sequence, we provide BLAST with a biochemical search image.

We expect that this approach will allow us to reduce the number of unannotated putative genes, while at the same time revealing insights into the mechanisms involved in protein evolution.