Analogical reasoning for knowledge discovery in a molecular biology database
Juergen Haas, Jeffrey S. Aaronson, G. Christian Overton · 1993
Genes are highly structured objects.Identifying genetic structure is experimentally costly and in general, is only partially determined in the lab.As such, genomic databases such as Gen-Bank are vastly underspecified.We describe various Case Based Reasoning (CBR) techniques, designed to take advantage of the homologous nature of genes, in order to discover regulatory features not present in the GenBank nucleic acid sequence database.We build a classification hierarchy reflecting the evolutionary relationship between genes.A sequence based CBR technique uses neighboring genes within the hierarchy as cases in order to predict the presence and location of regulatory features not listed in GenBank.Using a grammatical model of gene structure, we describe each gene with an instance grammar, and induce descriptive grammars of gene classes from these instance grammars.Descriptive grammars then serve as cases that enable us to identify regulatory features not present within the instance grammars.The above techniques recovered 92% of the regulatory features with an error rate of less than 2% for our data set. 1