Clustering mutational spectra via classification likelihood and Markov chain Monte Carlo algorithms

We have analyzed a set of 39 mutational spectra of the supF gene that were generated by different mutagenic agents and under different experimental co nditions. The cluster analyses was performed using a newly developed cluste ring procedure. The clustering criterion used in the procedure was develope d by applying the classification likelihood approach to multinomial observa tions. We also developed a Gibbs sampling-based optimization procedure that outperformed previously developed methods in a comparative simulation stud y. The results of the cluster analysis showed that our clustering procedure was able to recreate natural grouping of the mutational spectra with respe ct to the characteristics of mutagenic agents used to generate them and wit h respect to experimental conditions applied in the process of generating s pectra. These results are an important confirmation of the relevance of mut ational spectra in characterizing mutagenic mechanisms of different carcino gens.