Subclass Approach for Mutational Spectrum Analysis

Proceedings of the Twentieth International Conference on Machine Learning, 1995

Volume

Proceedings of the Twentieth International Conference on Machine Learning, 1995

Contents

Analysis and comparison of mutational spectra represents a burning question in molecular biology. We report an algorithm based upon the SEM subclass approach (SEM - stochastique, estimations, maximizations). Any real mutational spectrum is regarded as a mixture of standard binomial distributions. The separation procedure is run by rounds. Each iteration includes simulation, maximization and estimation. The algorithm has been checked on random spectra with the preset parameters and on real mutational spectra. As has been shown, any real mutational spectrum can be represented as a mixture of two and more binomial distributions, of which one contains hotspots of mutation.