A Classification-Driven Likelihood Ratio Method for Familial DNA Testing

Background: Familial DNA search relies on statistical methods to infer genetic relationships, often using the likelihood ratio (LR) as the test statistic. This approach involves calculating the probability of observed DNA profiles under various scenarios, traditionally assuming a uniform genetic background for all individuals. However, this assumption is unrealistic in practice due to population substructure, where allele frequencies differ among subpopulations. To address this, modified LR-based approaches have been proposed, including methods such as calculating likelihood ratios using average allele frequencies (LRLAF) or the maximum, minimum, and average likelihood ratios (LRMAX, LRMIN, and LRAVG). Despite these developments, further advancements are needed to improve the testing of genetic relationships in the presence of subpopulation structures. Results: We propose a novel LR-based statistic designed to account for population substructure by incorporating a classification step to address nuisance parameters. Specifically, two DNA profiles with unknown subpopulation origins are simultaneously classified into one group. Following this classification, the likelihood ratio (LRCLASS) is applied. Our analysis demonstrates that the proposed LRCLASS statistic, when paired with Naive Bayes classification, exhibits higher statistical power than existing methods for testing full-sibling relationships in the Thai population. Conclusion: The proposed LRCLASS statistic enhances the power of familial DNA searches by effectively addressing population substructure through a novel classification-based approach. This method provides a robust alternative to existing LR-based strategies and underscores the potential for integrating advanced classification techniques into genetic relationship testing.