Unsupervised Learning of Multiple Motifs in Biopolymers Using Expectation Maximization
Machine Learning - Special issue on applications in molecular biology
| Hi-index | 0.00 |
The identification of motifs in the gene promoters is a critical step in the delineation of the genetic regulatory framework of an organism. In this paper, a new linear mixed model is introduced. This model is a combination of the conventional Position Weight Matrix (PWM) model and a novel Mutual Information (MI) model. PWM can contain individual position frequencies whereas MI can reflect pair wise relation between positions. A training stage is carried out to determine the weight of each model. After that this trained model is embedded into a Gibbs sampling algorithm for motif discovery. After analyzing a set of DNA sequences using this program, putative motifs are gained and compared with experimental verified motifs as well as other popular motif finding software. Results show that this new mixed model can improve motif discovery accuracy to some extent.