Resumen
One of the main limitations of the typical hidden Markov model (HMM) implementation for gene structure identification is that a single structure is identified on a given sequence of genomic data?i.e., identification of overlapping structure is not directly possible, and certainly not possible within the confines of the optimal Viterbi path evaluation. This is a huge limitation given that we now know that significant portions of eukaryotic genomes, particularly mammalian genomes, are alternatively spliced, and, thus, have overlapping structure in the sense of the mRNA transcripts that result. Using the general meta-state HMM approach developed in prior work, however, more than one ?track? of annotation can be accommodated, thereby allowing a direct implementation of an alternative-splice gene-structure identifier. In this paper we examine the representation of alternative splicing annotation in the multi-track context, and show that the proliferation on states is manageable, and has sufficient statistical support on the genomes examined (human, mouse, worm, and fly) that a full alt-splice meta-state HMM gene finder can be implemented with sufficient statistical support. In the process of performing the alternative splicing analysis on alt-splice event counts we expected to see an increase in alternative splicing complexity as the organism becomes more complex, and this is seen with the percentage of genes with alt-splice variants increasing from worm to fly to the mammalian genomes (mouse and human). Of particular note is an increase in alternative splicing variants at the start and end of coding with the more complex organisms studied (mouse and human), indicating rapid new first and last exon recruitment that is possibly spliceosome mediated. This suggests that spliceosome-mediated refinements (acceleration) of gene structure variation and selection, with increasing levels of sophistication, has occurred in eukaryotes and in mammals especially.