NOC It : A computational method to infer the number of contributors to DNA samples analyzed by STR genotyping

Probabilistic characterisation of baseline noise in STR profiles

“…The Gaussian distribution class, which is commonly used to describe noise, gives the worst fit. This suggests replacing Gaussian noise models, as employed, for example, in [17], with those based on log-normal distributions in order to enhance the accuracy of deductions in the analysis of mixtures.…”

“…Taking the idea of the continuous methods further, it is consequent to remove the analytical threshold as well, and to consider the information in the peaks that were previously below the analytical threshold. The potential of this approach is illustrated in [17], where the authors present an algorithm to determine the number of contributors to a DNA mixture. They compare the maximum allele count method to their method, which uses no threshold, and show that the latter results in increased accuracy for samples with low DNA input mass.…”

“…Noise in STR profiles is commonly modelled via Gaussian distributions [18,17], though log-normal distributions have also been suggested [6]. In [17], a Gaussian noise model is used for determining the likelihood that a given number of individuals contributed to a mixed sample.…”

“…In [17], a Gaussian noise model is used for determining the likelihood that a given number of individuals contributed to a mixed sample. Although utilisation of the Gaussian model to describe noise provides improved identification over previous techniques, the authors do not provide analysis independent of determining the most likely number of contributors to confirm its appropriateness.…”

The Probabilistic Genotyping Software STRmix: Utility and Evidence for its Validity

“…Under this process, a NoC is assigned to the profile based on the number of allelic peaks and their heights, often after consideration of artifacts. While most constrained NoC determinations are performed manually, there are software tools available to assist if needed, such as NOCIt based on Monte Carlo methods, PACE based on machine learning, and methods using Bayesian networks and maximum likelihood .…”