Reference-based gene model prediction on DNA contigs

摘要

This paper presents an algorithm for constructing multiple gene models on a set of contigs of a large genomic clone. The algorithm first uses pattern recognition-based methods to locate exons or partial exons in each contig, and then applies protein homology or EST information from the databases, as reference models, to parse the predicted exons into gene models. In the phase of gene model construction, the algorithm uses a unified framework for genes ranging from situation with homologous proteins/ESTs to no homologous protein/EST in the database. By exploiting protein homology or EST information, the algorithm is able to (1) parse exons into multiple gene models over a set of DNA contigs (possibly unoriented and unordered); (2) remove falsely predicted exons; and (3) identify and locate exons missed by the initial exon prediction.