Characterizing the genetic structure of a forensic DNA database using a latent variable approach

Area of Science:

• Forensic Genetics
• Population Genetics
• Statistical Modeling

Background:

• Forensic DNA databases are crucial but challenging to model accurately due to unknown ethnic compositions.
• Existing methods often rely on small, potentially unrepresentative population surveys.

Purpose of the Study:

• To develop a method for estimating allele frequencies and subpopulation proportions directly from offender DNA databases.
• To create a representative model of a forensic DNA database using a latent variable approach.

Main Methods:

• A latent variable model was developed to estimate allele frequencies and subpopulation proportions from the database itself.
• The Expectation-Maximization (EM) algorithm was used for parameter estimation.
• The model was applied to the Dutch offender database (2014) with approximately 180,000 profiles.

Main Results:

• Three subpopulations were found to adequately represent the heterogeneity within the Dutch offender database.
• The model accurately predicted false leads in database searches, including familial searching and mixture analysis.
• Analysis of partial matching and validation of the Balding-Nichols formula with a specific theta value were performed.

Conclusions:

• The latent variable approach provides a reliable and data-driven method for modeling forensic DNA databases.
• This approach enhances the accuracy of forensic genetic analyses and database search predictions.
• The findings support the conservative nature of current Dutch match probability estimation practices.