An important foundation for the advancement of both basic and applied biological science is correct annotation of protein-coding gene repertoires in model organisms. Accurate automated annotation of eukaryotic gene structures remains a challenging, open-ended and critical problem for modern computational biology.;The use of extrinsic (homology) information has been shown as a quite successful strategy for this task, though it is not a perfect solution, for a variety of reasons. More recently, gene prediction methods leveraging information present in syntenic genomic sequences have become favorable, though these too, have limitations.;Identifying genes by inspection of genomic sequence alone thoroughly tests our theoretical understanding of the gene recognition process as it occurs in vivo, and where we encounter failure, excellent opportunities for meaningful research are revealed.;Therefore, the continued development of methods not reliant on homology information---the so-called ab initio gene prediction methods---should help to more rapidly achieve a comprehensive understanding of gene content in our model organisms, at least.;This thesis explores the development of novel algorithms in an attempt to advance the current state-of-the-art in gene prediction, with particular emphasis on ab initio approaches.;The work has been conducted with an eye towards contributing open source, well-documented, and extensible software systems implementing the methods, and to generate novel biological knowledge with respect to plant taxa, in particular.