Statistical interpretation of DNA from forensic evidence in crime
cases may be computationally extremely demanding if the sample
contains DNA from many people. While the forensic and statistical
interpretation of the DNA sample concerns the DNA profiles of the
people contributing to the sample, we may only observe the mixed
signal of DNA components in the sample. The large discrete state space
of the DNA profiles is the common root cause of a high computational
complexity for all of the statistical models used for DNA
interpretation in casework.

I will explain a computational approach based on well-established
algorithms for graphical models -- one immediate advantage being that
implementations are available in standard software. All computations
are exact while still efficient enough to allow the interpretation of
the more complex DNA samples occurring in practical casework.

Many of the ideas are more generally applicable, for instance in
computing the likelihood in a model with discrete latent variables.