More recently, Anders Krogh, Saira Mian, and David Haussler at UC Santa Cruz have built a hand-tuned Hidden Markov Model [7] that recognizes genes in E.Coli DNA. Their design includes a special HMM trained to recognize promoter and translational initiation signals as well as look for regions of gene content using codon usage measure. Snyder and Stormo in CU Boulder [14] have used machine learning to predict coding regions using a neural network. Their program learns the best way to find exon/intron boundaries in human DNA. A good review of the concepts behind DNA parsing and HMMs is by David Searls [13].