MINING SPECIES-SPECIFIC SUBSEQUENCES IN BACTERIA TRANSCRIPTION TERMINATORS

摘要

Transcription Terminators (TT) play an important role in bacterial RNA transcription. Some bacteria are known to have Species-Specific Subsequences (SSS) in their TTs, which might contain useful clues to bacterial evolution. Given DNA sequences for known TTs, how to effectively find the SSS is an interesting yet not well-studied computational problem. In this paper, we present a three-step method to identify the SSS by using Support Vector Machines (SVM). Firstly, we find out all frequent subsequences by using generalized suffix trees. Secondly, we use the subsequences as features to vectorize the DNA sequences and trains a SVM classifier. Finally, we output from the SVM classifier the SSS based on a defined measure of subsequence specificity. Our experiments show that the SSS are very close to the known SSS. We conclude that our method exhibits a novel application of classification to biology and is applicable to similar problems.