A platform for applying multiple machine learning strategies to the task of understanding gene structure

Describes a system to support multiple machine learning strategies applied to the analysis of gene structure in general and the fine-structure of gene regulatory regions in particular. The focus is on two related tasks. The first is to take an overly generalized description of gene structure, represented as a formal grammar, and generate a detailed fine-structure description of a gene family (in this case the, beta -hemoglobin family); this amounts to a restricted form of grammar induction. The second task is to identify patterns in the sequence of uncharacterized DNA that specify functional regions involved in gene regulation. Because too few properly classified training examples are usually available to support statistical induction of pattern descriptors, the authors have instead used a variant of case-based reasoning to identify patterns in uncharacterized DNA by comparison to analogous cases in well-characterized DNA.<
>