A New Specification to Gene Signals Sensors by Neural Self Organizing Feature Map (SOFM)

Two different paradigms by the goals in gene-finding research have been recognized: 1) to offer computational aid in the annotation of the large volume of genomic data and 2) to provide a computational model helpful in elucidating the mechanisms involved in transcription, splicing, polyadynalation and other important processes on the pathway from genome to proteome . New findings in gene regulation appear to focus a new interest in the latter paradigm approaches. Therefore, a single weight matrix for the genomic patterns consensus scoring can be substituted by SOFM clusters matrices. This should result in the detection improvement of gene functional sites or signals, and therefore a gain evaluation across the known Burset\´s and Guigo\´s collection of the genes of 570 vertebrates is provided by a percentile measure on an exemplary site detection statistics. Such an improvement is important in both the "extrinsic" and "intrinsic" approaches. In the "signal" case of the latter, a demand is addressed for a neural approach which translates into likelihood scoring. This includes, but is not limited to, applications with HMM (hidden Markov model) derived gene finders. The approach is also scalable into a clusters-based solution to genes recognition with the capability of integrating DNA-string-contained knowledge in a novel way