Bayesian regression analysis of stutter in DNA mixtures

Abstract

Probabilistic genotyping methods use a hierarchical probability model in deconvolution of DNA mixtures. The parameters of the model, including the stutter which are required to calculate the expected values of peak heights, are estimated in the validation process. Linear modeling of stutter, as a common artifact in DNA genotyping, has been reported previously. The typically right-skewed error distribution and non-negativeness of stutter to its allele peak heights ratios make generalized linear models preferable, especially Bayesian analogs, which allow even more flexibility. In this paper, we show how such models can be fitted and applied with the aid of Markov chain Monte Carlo methods.

Keywords:

• Bayesian methods
• DNA genotyping
• generalized linear models
• Markov chain Monte Carlo methods
• stutter

MATHEMATICS SUBJECT CLASSIFICATION:

• 62F15; 62J12

Additional information

Funding

The authors wish to thank Rebecca Klein for the amplification data. R. Alaeddini was supported by the US National Institute of Justice under Grant [number 2016-DN-BX-0082].