The Twinscan program predicts genes in a manner similar to Genscan, except that Twinscan takes advantage of genome comparison to improve gene prediction accuracy. In the version of Twinscan used to generate this track, intronless copies of known genes are masked out before gene prediction, reducing the number of non-processed pseudogenes in gene models. More information and a web server can be found at http://genes.cs.wustl.edu/.
The Twinscan algorithm is described in Korf I, Flicek P, Duan D, and Brent MR (2001), "Integrating genomic homology into gene structure prediction", Bioinformatics 17:S140-148.
Thanks to Michael Brent's Computational Genomics Group at Washington University St. Louis for providing these data.