Importance sampling allows H_d true tests of highly discriminating DNA profiles

Area of Science:

• Forensic Genetics
• Computational Biology
• Statistical Modeling

Background:

• H_d true testing assesses DNA interpretation system performance by simulating non-donor profiles.
• A key validation criterion is that the average likelihood ratio (LR) should approximate one.
• Current methods face computational limitations due to the large number of simulations required, especially for profiles with high discrimination power.

Purpose of the Study:

• To address the computational limitations of H_d true testing.
• To introduce and explain the application of importance sampling in forensic genetics for H_d true tests.
• To demonstrate that importance sampling enables H_d true testing for all DNA profiles.

Main Methods:

• Simulating DNA profiles of non-donors under two propositions (contribution vs. non-contribution).
• Calculating likelihood ratios (LRs) to assess model performance.
• Employing importance sampling to enhance the simulation of rare events (non-donor contributions).
• Adjusting simulation results to correct for the bias introduced by importance sampling.

Main Results:

• Importance sampling significantly reduces the computational burden of H_d true tests.
• This technique allows for the simulation of rare events to occur more frequently.
• The method recovers all diagnostic values of interest by adjusting for simulation bias.
• H_d true testing becomes applicable to all DNA profiles, irrespective of their discrimination power.

Conclusions:

• Importance sampling is a powerful, yet underutilized, technique for H_d true testing in forensic genetics.
• This method overcomes the computational limitations previously hindering comprehensive software validation.
• The application of importance sampling ensures robust H_d true testing capabilities across diverse DNA profiles.