Performance of GENSCAN

Performance of GENSCAN has been compared to that of other computer programs written for gene finding. The Burset/Guigo set of 570 vertebrate multi-exon gene sequences [4] was used as the test set. The results (as reported in [4]) have been quite encouraging. Both at the nucleotide level as well as the exon-level, GENSCAN's accuracy has been significantly better than other programs.
Important features of GENSCAN include:

• Identification of complete intron/exon structures of a gene in genomic DNA.
• Ability to predict multiple genes and to deal with partial as well as complete genes.
• Ability to predict consistent sets of genes occurring on either or both strands of the DNA.
• Ability to predict both optimal annotation and sub-optimal exons.

  

Figure: Accuracy of GENSCAN for different signal and exon types [3].

 

Figure: GENSCAN results [3].

Although the results are good they are still not good enough for massive gene finding. GENSCAN has 80% chance for detecting an exon. If a gene has more than one exon the probability of correctly detecting all of them declines rapidly. On the positive side a more permissive figure of merit, the prediction per bp, is over 90%.