Mutational signatures are probability vectors over the different mutation types that represent specific mutational processes. In cancer genomics, the mutational profile of a patient is a mixture of such mutational processes. Mutational signatures are usually derived using non-negative matrix factorization (NMF). To extract the mutational signatures we have to choose an error model for the observed mutational counts, which determines the underlying distributional assumption of the data. In most applications, the mutational counts are assumed to be Poisson distributed, but this is often overdispersed and leads to an overestimation of the number of signatures. We introduce a different error model where the mutational counts follow a Negative Binomial distribution and an approach to accurately estimate the number of signatures. Furthermore, we extend the definition of a mutation type and include flanking nucleotides further away from the base substitution. By parameterizing the signatures and using mutational opportunities we show that our method provides more robust and interpretable signatures also in extended nucleotide contexts.