Knowledge Acquisition from Amino Acid Sequences byMachine Learning System BONSAI Authors

We present a machine learning system, called BONSAI, for knowledge acquisition from positive and negative examples of strings, and report some experiments on protein data using the PIR and GenBank databases. This learning system is constructed with an algorithmic learning theory for decision trees over regular patterns, which is newly developed for this work. As a hypothesis, the system tries to nd a pair of a classi cation of symbols called an alphabet indexing and a decision tree over regular patterns, which classi es given examples with high accuracy. Through the experiments, the system discovered very simple hypotheses that exhibit important knowledge about transmembrane domains and signal peptides.