Internal validation of the fully continuous model in EuroForMix for its implementation in routine forensic DNA profiling

Despite being increasingly more common in the forensic routine, DNA mixtures still present an important roadblock in the analysis of genetic evidence. The International Society for Forensic Genetics (ISFG) recommends the use of the Likelihood Ratio (LR) in the statistical evaluation of the evidence for the contribution of a certain Person of Interest (POI). Numerous mathematical models have been developed for this purpose, which can be classified according to the information included: i) binary models only use the allelic information, ii) qualitative or semi-continuous models also accommodate drop-out and drop-in probabilities and iii) quantitative or fully continuous models include additionally allelic signal intensity (peak heights or read coverage). Before being implemented in the forensic routine, any software that supports a form of these models must be tested in different scenarios to perform an internal validation. In this work we present the results of our internal validation of the fully continuous model present in EuroForMix (EFM) and DNAStatistX (used as control of EFM in this study), comparing its performance to the semi-continuous model in LRmix Studio. To do so, 59 artificial samples were prepared from 8 unrelated contributors in mixtures of 2, 3 and 4 contributors; besides, 19 tests from 7 real cases were run in both software. The following analyses were conducted: i) scrutiny of contributors in single-source serially diluted samples; ii) estimation of the percentage of contribution to the mixture; iii) comparison of results obtained in artificial mixtures for H1-True and H1-False hypotheses; iv) effect of sensitivity analysis in the final result; v) discrimination power of each model; vi) the behaviour of the Automatic Model Search option in EuroForMix and vii) comparison of results obtained in real casework analyses. Our results show that EFM returns higher values than LRmix Studio both in single-source and mixed scenarios. The fully continuous model has demonstrated a neutral behaviour, returning inconclusive results when genetic evidence is scarce. If a reporting threshold is applied EFM is especially affected, with a higher proportion of negative log(LR) results turning inconclusive in comparison to LRmix Studio. Overall, EuroForMix has shown a correct performance, returning reliable results and reducing the subjectivity of the analysis.