Applying machine learning techniques to DNA sequence analysis. Final report

摘要

The authors created computational tools for finding genes in anonymous DNA, in close collaboration with the NIH-funded project of F. Blattner whose goal is to determine the complete DNA sequence of E. coli. They primarily developed a machine learning (ML) system that modifies existing knowledge about specific types of biological sequences. Specifically, the KBANN algorithm maps inference rules about a given recognition task into a knowledge-based neural network. Neural network training techniques then use the training examples to refine these inference rules. In addition to the KBANN work, the authors applied neural networks to the task of recognizing coding regions in E. coli. E. Uberbacher, R. Mural, and their group at the Oak Ridge National Laboratory installed the resulting trained neural network as the GRAIL-E email server, a beta-test version of their GRAIL system to be used for analyzing E. coli sequences. They also integrated this trained neural network into the graphical interface to the FIND-IT system. The basic FIND-IT reports alignments to known proteins, while the neural network suggests possible coding regions in unmatched DNA segments. The authors also developed software tools that perform sequence alignments in the presence of frameshift errors.