Utilising allelic dropout probabilities estimated by logistic regression in casework.

Area of Science:

• Forensic Science
• Genetics
• Statistical Modeling

Background:

• DNA profile interpretation in forensics often requires estimating the probability of allelic dropout.
• Existing logistic regression methods model dropout using locus-specific constants, with variations based on template quantity proxies.

Purpose of the Study:

• To evaluate a novel variant of a logistic regression model for DNA dropout probability estimation.
• To compare the performance of this variant against existing methods in a casework-realistic training/testing scenario.

Main Methods:

• Developed and tested two established logistic regression models for dropout probability.
• Introduced and evaluated a variant model with constants constrained to be identical across all loci.
• Trained model constants on one dataset and tested predictive performance on an independent dataset.

Main Results:

• The novel variant model, with locus-independent constants, demonstrated superior performance in predicting dropout events compared to the other two methods.
• The enhanced accuracy of the variant suggests its utility in real-world forensic casework.

Conclusions:

• A logistic regression model variant with shared constants across loci is a more robust and accurate tool for estimating DNA dropout probabilities.
• This approach mitigates issues arising from locus-specific variations in amplification efficiency, improving casework reliability.