Machine Discovery of a Negative Motif

Setsuo Arikawa, Satoru Kuhara, Satoru Miyano, Ayumi Shinohara, Takeshi Shinohara
1991 Genome Informatics Series  
Traditional approaches to motif-searching in proteins are to find subsequences common to functional domains by various alignment techniques. However, a machine learning system, which we developed using a new concept called a decision tree over regular patterns, discovered that significant motifs do not always locate in the functional domains but in the outside. Experiments were made for identifying hydrophobic transmembrane domains. Two consecutive polar amino acids that were found in
more » ... embrane domains can identify transmembrane domains with 90% accuracy for all data in PIR database. This machine learning approach may provide a new method for discovering motifs.
doi:10.11234/gi1990.2.62 fatcat:uhfbj2c4x5htnp5j36ls7dgub4