A New DSP-Based Measure for CPG Islands Detection

Identification of DNA regions that are rich in the CpG pattern is important because such segments are resistant to methylation and are typically associated with genes that are frequently switched on. In this paper, we propose a DSP based approach to detect such regions. In specific, we first derive two new Markov chain models for CpG and non-CpG regions respectively and then design a number of FIR filters to generate a filtered Likelihood Ratio Test (LRT) measure. Our simulations show that the joint design of the probabilistic models and the digital filter produces significantly better results than previous work