Context dependent DNA substitution models

Abstract

When modelling the substitution process of DNA sequence, most previous work assumes that nucleotides are independent of each other. However, empirical evidence suggests that the context dependent model is more accurate to this problem. Thus, there is a great demand for statistical approaches for context dependent substitution models. We propose a general context dependent substitution model and investigate a special case, a two-flanking sites model. To reduce the number of parameters in the model, we derive two sub-models by clustering the substitution matrices. Moreover, we develop two methods to estimate the parameters in our models. Our modified Parsimony method can handle context dependent models. Using results from Parsimony method as initial values for optimization, our Maximum Pseudo-likelihood method can achieve more accurate results. We also build a simulation process, which can generate unlimited testing data. Experiments have been done on the simulation data and real data. The experiments show the success of our proposed models and methods.