Historically, forensic DNA analysis has centered on STR data. STR profiles have become the gold standard for human identification. With increasingly sensitive chemistries, even tiny traces of DNA can create complicated profiles, particularly for evidence samples that contain the DNA of more than one individual. Therefore, most research has focused on making sense of complex STR mixtures.

Additionally, most crime labs still haven’t adopted SNP technologies. The high price of sequencing technologies and the previously unexplored utility of SNP data in forensic casework have kept labs focused on traditional STR methods. However, the growing public interest in methods that use SNP data, such as forensic genetic genealogy, combined with the need for tools that can extract more genetic information from highly degraded samples, are motivating crime labs to consider adopting SNP technologies. As sequencing tools and forensic SNP kits become more affordable and widely accepted, we are entering a new era, one where SNP technologies are more commonly used in casework. And with that comes a big question: How do we interpret SNP mixtures?

This challenge hasn’t received much attention until now. It is worth noting that probabilistic methods for interpreting STR mixtures took decades of research before they were adopted in forensic casework. Now is the time to apply that same foresight to SNP mixture data. Research conducted today will shape the future of SNP forensic data analysis.

The first step of this project involves the development of a bioinformatics framework for simulating SNP data. Our approach takes advantage of the binomial nature of SNPs, where most DNA locations have only two allele options, for example, allele A or allele B. Similar to how peak heights in STR data reflect the amount of DNA for each allele, SNP data is represented by read counts, which indicate the amount of DNA detected. In SNP mixture data, the number of reads for each allele depends on the number of contributors in the sample, their corresponding genotypes, and their mixture proportions - which is a measure of how much DNA each individual contributes to the mixture.

Using these variables, we have simulated SNP mixtures and are currently applying a likelihood model designed to infer the alleles of each contributor in two-person mixtures. Our model will be tested with simulated mixture data that mimics real forensic conditions by incorporating parameters that can account for stochastic effects and sequencing errors.

One current limitation is that our framework does not account for population-specific allele frequency differences, which can influence genotype distributions and affect mixture interpretation. To address this, our next step is to integrate data from the 1000 Genomes Project to evaluate mixtures across diverse populations and assess the accuracy of our model using data that better represents human genetic diversity.

Our probabilistic model for deconvoluting two-person mixtures shares similarities with existing STR interpretation methods used in crime labs, as both rely on likelihood-based approaches to evaluate which hypotheses are most consistent with the observed data. However, a key difference is that current STR mixture interpretation produces likelihood ratios that compare two competing hypotheses, whereas our approach focuses on inferring the most likely contributor genotypes given the mixture data. Therefore, the result is not in the form of a likelihood ratio, but rather an individual’s most probable genotype at a given number of SNPs. Another major distinction, though not specific to the method itself, is the genetic architecture of STR versus SNP data. STR loci are multi-allelic, which aids mixture interpretation, while SNPs are bi-allelic, offering only two allele options. This reduced variability makes SNP mixture interpretation more challenging. Additionally, the scale of analysis differs significantly because STR interpretation generally involves 22-26 loci, whereas SNP-based analysis requires evaluating thousands of loci. This makes computational efficiency a critical requirement for SNP mixture interpretation tools.

This question makes me think about the balance between creativity and practicality in research. In research spaces, there’s often room for infinite innovation and creativity when proposing solutions to problems. However, forensic research requires solutions that are not only novel but also feasible and applicable in casework. I worked as a crime lab DNA scientist for 5 years, and I carry that experience into my research. Having been in their position, I approach research questions and solutions from the perspective of casework scientists. . This perspective ensures that while I explore creative solutions, I develop methods that can be implemented in crime labs and can be explained during court testimony – a key step in ensuring that scientific results from crime scene evidence are admissible in court. In short, innovation is essential, but it must be practical and have procedural continuity for it to be impactful in real-world casework.

I don’t anticipate that implementing SNP mixture interpretation will significantly change day-to-day forensic workflows in the near term. This is because most forensic workflows remain centered on STR data and STR mixture interpretation. Crime scene evidence will continue to be processed with the primary goal of generating STR profiles, which depending on the case, can be uploaded to the Combined DNA Index System (CODIS) to provide investigative leads. SNP mixture interpretation will primarily be critical in cases where all traditional forensic resources have been exhausted and the evidence qualifies for forensic genetic genealogy approaches. In those instances, mixtures could be processed through the SNP mixture workflow, and the resulting data could then be used to support genetic genealogy investigations. However, in the future, once more crime laboratories begin adopting sequencing technologies, SNP mixture interpretation will become more common, having the potential to significantly transform day-to-day forensic workflows.

This question reminds me of my time in casework, when investigative genetic genealogy (IGG) approaches first gained attention after their 2018 success in identifying the Golden State Killer - a case that had puzzled investigators in California for decades. That breakthrough sparked a surge of interest from investigators, district attorneys, and detectives eager to determine whether their cold case samples were eligible for IGG. These requests were typically outsourced, meaning evidence samples were sent to external agencies for sequencing and genealogy searchers. Eligibility, however, depended heavily on sample quality, particularly whether the sample was single-source or a mixture. If it was a mixture, the sample was deemed ineligible. I remember thinking about how this excluded so many cases where IGG could have been useful and possibly the last resort for generating investigative leads. This was especially concerning to me because, due to the increased sensitivity of modern forensic kits, most evidence samples result in mixture data. Unlocking SNP mixtures for FGG eligibility would dramatically expand the number of samples that qualify for genetic genealogy, creating new opportunities to apply advanced methods to cold cases and potentially provide investigative leads in cases that have remained unresolved for decades.

SNP deconvolution could be especially transformative for cold cases, regardless of the type of crime. However, initial applications should focus on high-yield, high-quality samples (e.g, blood, saliva, etc.). Degraded or low-level samples introduce additional interpretation challenges that must be evaluated in mock case samples before implementing this approach in routine casework.

My decision to pursue a PhD after working in the crime lab was one of the most difficult yet pivotal decisions in my career. What made it particularly challenging was enrolling in a program focused on human medical genetics, even though my long-term goal was forensic research. I worried that the steep learning curve in medical genetics would make transitioning into a PhD program difficult, and that the knowledge I gained in medical genetics might not translate back to forensics. However, I had one goal in mind: to expand my knowledge hoping that anything I’d learn could one day help me address the gaps and limitations in forensic genetics that I observed while working casework. Thankfully, I quickly realized that research questions in both medicine and forensic science rely on the same fundamental genetic and statistical principles, differing primarily in their application. Looking back, I am grateful that I took this risk because it expanded my expertise in genetics, bioinformatics, and statistics. I’ve been exposed to methods routinely used in human medical and population genetics. In fact, the knowledge I gained through my PhD sparked the idea of addressing the SNP mixture interpretation problem, and that knowledge was the foundation for the NIJ work I am doing today.

My advice is this: If there’s something you think about often while at work, perhaps you have an idea that could really make a difference in casework, you should know that your idea matters, your experience is invaluable, and you are capable of pursuing it. It’s easy to assume that translational research strictly requires a PhD, but that’s not necessarily the case. I’ve seen incredible forensic scientists who have conducted impactful research in their own labs. I would encourage forensic scientists to start sharing their ideas with leadership to explore grant opportunities to fund research projects. This can create space for scientists to step away from casework temporarily and dedicate time to research that advances both their laboratory and the broader forensic science community.