Genome-Wide Forensic DNA Analysis

The ability to identify individuals by comparing DNA profiles from questioned biological samples to those of presumptive relatives is an application of central importance to forensics. Likewise, accurate DNA-based kinship analysis is a biometric tool of great potential importance to immigration and national security. Comparison to family references is a primary method for missing persons or disaster victim identification (DVI; where reliable direct references are commonly not available). However, its utility is greatly constrained by the need for multiple immediate relatives to achieve a high certainty of identification. Lack of close family reference samples has been a key limitation for the identification of victims of major US disasters and in the US national missing persons program. The ability to detect distant relationships between people with high certainty would resolve a fundamental weakness in forensic DNA analysis and would have applications in security and immigration biometrics, identification of unknown human remains, and lead generation in criminal investigations. We have developed a computational method (ERSA) that uses high-density SNP genotypes to link relatives as distant as third cousins with high confidence. The main obstacle limiting the application of ERSA to forensic DNA is that it requires high-density, high-quality SNP genotype data, and such data cannot currently be generated from many forensic DNA samples: they are frequently too fragmented, chemically damaged, inhibited, or limited in quantity. Techniques for repairing and amplifying such DNA are steadily improving, but the data collected from even modestly limited or damaged DNA are incomplete and error prone. Since ERSA and other available relationship inference tools were designed for high-quality data, they are not applicable to most forensic problems. The crucial advantage of high-density genotyping methods is the vast number of loci they interrogate. This enables them to collect very large amounts of accurate information even with high per-locus failure rates that may be observed with forensic DNA samples. Here we propose to determine the breadth of applicability, cost-effectiveness and robustness of ERSA and other relationship estimation methods in forensic contexts, especially on incomplete and error-prone data; to adapt ERSA to handle such data better; and to explore error-insensitive methods for inferring relationships from data sets that are so sparse or error-prone that existing methods cannot be applied at all. To guide our analyses and development, we will characterize the types and frequencies of errors that appear in high-density SNP data by genotyping replicated DNA samples covering the range from high quantity and quality to limiting quantities of severely fragmented DNA. We will then use publicly available high-density SNP data sets in conjunction with empirically-based error models to simulate genomic data representing forensically relevant problems and test ERSA and other methods on these data. The results will guide us as we adapt ERSA to error-prone data and explore error-insensitive approaches. The data we generate will be made publicly available and the computational tools (ERSA, its modifications, adaptations of existing methods, and novel relationship estimation methods) will be either publicly available or free for research use. We will publish our results in peer reviewed forensics journals, present them at forensics conferences, and provide them to the NIJ in interim and final reports. This research has the potential to introduce fundamentally new data analysis methods to forensic DNA science and to greatly expand the potential for the identification of missing persons, victims and suspects in criminal investigations, and disaster victims by linking them with distant kin. ca/ncf