GeneWise and Genomewise

Abstract

We present two algorithms in this paper: GeneWise, which predicts gene structure using similar protein sequences, and Genomewise, which provides a gene structure final parse across cDNA- and EST-defined spliced structure. Both algorithms are heavily used by the Ensembl annotation system. The GeneWise algorithm was developed from a principled combination of hidden Markov models (HMMs). Both algorithms are highly accurate and can provide both accurate and complete gene structures when used with the correct evidence.

Figure 1

The merging process between the gene prediction model and the homology model (written out in the top panel) and the merged model (lower panel). The large black arrows indicate that between each set of four states, all possible transitions exist. This model will be later pruned to a small set of states: See the text for more details.

Figure 2

The GeneWise21:93 algorithm. The circles represent states, and the arrows between them transitions. Black transitions are standard protein transitions, red transitions are frameshifting transitions, and green transitions are intronic transitions. Introns are each built of three states, listed at the bottom of the figure.

Figure 3

The GeneWise6:23 algorithm. The circles and rectangles represent states, and the arrows between them transitions. For simplicity the three intron states are gathered into one node; in each case there is a unique transition state from either Match or Insert into each of the intron states (each state representing a different phase).