学会活動

2010年2月8日(月)に「第5回JSBiバイオインフォマティクス学会北海道地域部会セミナー」を北海道大学情報科学研究科にて開催致します。
今回は、京都大学・化学研究所の馬見塚拓先生をお招きし、木パターンマイニングに関する研究成果を講演して頂きます。奮ってご参加下さい。

日時: 2月8日 15:00～16:00
会場: 情報科学研究科棟 A11講義室
講師：京都大学化学研究所 バイオインフォマティクスセンター
馬見塚　拓 教授
参加費：無料

講演タイトル：
Mining patterns from trees - Probabilistic model-based approach

要旨：
Trees are a typical example of semi-structured data that appear in many applications, including text, web and molecular biology.
Specifically the motivated application of this talk is carbohydrate sugar chains (or glycans), which are important biological components as well as can be labeled ordered trees in a computer science sense. This talk will start with a brief introduction of glycans or labeled ordered trees, being followed by the description on hidden Markov model (HMM) which is a standard probabilistic model for
handling patterns in time-series data or sequences. Then a
probabilistic framework is presented for mining patterns from trees. Models in this framework are a reasonable extension of a vareity of probabilistic models, including HMM. The learning scheme of the models, being based on the EM (Expectation and Maximization) algorithm, is also a natural extension of those for various probabilistic models, including the Baum-Welch (or Forward-Backward)
algorithm of HMM. The performance of the proposed scheme was
empirically demonstrated in two different ways. First the predictive performance was measured in a two-class classification manner using both synthetic and real datasets, showing the advantage over existing approaches in learning from trees. Secondly the performance on real
data was evaluated from a biological viewpoint, confirming existing knowledge in glycobiology.