A Stochastic Approach to Genetic Information Processing

Abstract

This paper stresses the importance of stochastic machine learning theory for analyzing genetic information such as protein sequences. It is commonly recognized that machine learning theory would play an essential role to extract important information from the enormous amounts of raw genetic information generated by biologists. However, it is also true that more flexible and robust learning methodologies are required to deal with divergence occurring on the genetic information. For this purpose, we adopt stochastic knowledge representations and stochastic learning algorithms and show their effectiveness with a stochastic motif extraction system. The system aims to extract stable common patterns conserved in some protein category. In the system, common patterns (stochastic motifs) are represented by stochastic decision predicates, and a genetic algorithm with Rissanen's minimum description length principle is used to select “good stochastic motifs” from the viewpoint of increasing prediction performance.