RFU derived LRs for activity level assignments using Bayesian Networks

Peter Gill; Øyvind Bleka; Ane Elida Fonneløp

doi:10.1016/j.fsigen.2021.102608

RFU derived LRs for activity level assignments using Bayesian Networks

Forensic Sci Int Genet. 2022 Jan:56:102608. doi: 10.1016/j.fsigen.2021.102608. Epub 2021 Oct 21.

Authors

Peter Gill¹, Øyvind Bleka², Ane Elida Fonneløp²

Affiliations

¹ Forensic Genetics Research Group, Oslo University Hospital, Oslo, Norway; Department of Clinical Medicine, University of Oslo, Oslo, Norway. Electronic address: peterd.gill@gmail.com.
² Forensic Genetics Research Group, Oslo University Hospital, Oslo, Norway.

PMID: 34735938
DOI: 10.1016/j.fsigen.2021.102608

Abstract

A comparative study has been carried out, comparing two different methods to estimate activity level likelihood ratios (LR_a) using Bayesian Networks. The first method uses the sub-source likelihood ratio (log₁₀LR_ϕ) as a 'quality indicator'. However, this has been criticised as introducing potential bias from population differences in allelic proportions. An alternative method has been introduced that is based upon the total RFU of a DNA profile that is adjusted using the mixture proportion (M_x) which is calculated from quantitative probabilistic genotyping software (EuroForMix). Bayesian logistic regressions of direct transfer data showed that the two methods were comparable. Differences were attributed to sampling error, and small sample sizes of secondary transfer data. The Bayesian approach facilitates comparative studies by taking account of sampling error; it can easily be extended to compare different methods.

Keywords: ALTRaPht; Bayesian Network; Direct transfer; EuroForMix; Evidence evaluation; Likelihood ratio (LR); Mixtures; Secondary transfer.

MeSH terms

Bayes Theorem
DNA Fingerprinting*
Humans
Likelihood Functions
Microsatellite Repeats*
Software