The ISHI agenda is live and includes great talks from amazing speakers! While the forensic community is a tight-knit group, we can always get a little closer, right? With that in mind, we interviewed our speakers to preview their presentations and to get to know them a little better outside of their work. We’ve been posting their responses in a feature we like to call Under the Microscope.
Today, we’re chatting with Jonathan Adelman and Michael Marciano, who will be co-presenting A Probabilistic Machine Learning-Based Assessment of the Number of Contributors in DNA Mixtures during the General Sessions on Wednesday, October 4th.
You mention in your abstract that the number of contributors in a given DNA sample is the most critical component of de-convoluting a mixture. Can you explain why this is for those who may not know?
Marciano: The foundation of DNA mixture interpretation is built upon a set of assumptions. The number of contributors is generally recognized as the most critical. Historically, the assumption of the number of contributors was the single most critical metric in standard manual methods of deconvolution because it sets the limit on the number of allele peaks that can be attributed to any one contributor. Current mixture interpretation methods such as probabilistic genotyping still rely on, or greatly benefit from, assuming the correct number of contributors. Incorrect assumptions may lead to lower likelihood ratios and challenges relating to inclusion/exclusions.
How do labs currently estimate the number of contributors in a DNA sample?
Marciano: The Maximum Allele Count method remains predominant in the community. Many laboratories have started using additional data such as the sample-wide allele count. Our opinions are based on my past experience in casework and have been developed through discussions with several laboratories.
How will using a machine-based approach impact a lab’s current workflow?
Marciano: Generally, the approach will be no different than using peak detection software. A standard laptop or desktop computer can be used. The machine learning-based approach is rapid—results can be obtained within seconds—and lends itself to a seamless transition into standard workflow.
How did you become interested/involved in this project?
Marciano: Simply put, it was born out of two individuals with complementary expertise and a white board. My friend and colleague, Jonathan Adelman, and I developed this idea in 2013. I have a background in molecular biology; I spent my first 5-6 years as a casework analyst at the Onondaga County Center for Forensic Sciences (Syracuse, NY). I later moved my career in the direction of applied research and began working at a defense contractor where I worked closely with Jonathan addressing applied non-human problem sets in the Department of Defense and the intelligence community. We joined the Syracuse University Forensic and National Security Sciences Institute and were given the freedom to explore new approaches to human genetic identity analyses. We were able to further formalize our machine learning approaches through a tangential project funded by National Institute of Justice. This was an example of scientific innovation born from interdisciplinary approaches.
Adelman: I was utilizing artificial intelligence and machine learning in several projects for the defense and intelligence communities when Mike first came to me and asked if the idea of using A.I. for mixture interpretation problems made sense. The biggest surprise for me in that first year of research – and it remains so today – is how little these computational approaches are being leveraged for forensic science problems even though their strengths and weaknesses often dovetail beautifully with key challenge problems in the domain, and I remain very interested in exploring how machine learning can solve forensic challenges.
What do you feel is the biggest challenge that forensics laboratories are facing today?
Marciano: In terms of interpretation, we have heard directly from many laboratories that the prediction of the number of contributors remains the biggest challenge. However, another major challenge is to acclimate scientists to new types of analytical and statistical methods, such as those used in probabilistic genotyping. While training programs are very well constructed and executed, the constant rigors of casework leaves little time to develop the appropriate expertise in a timely fashion, particularly in smaller laboratories. I do recognize, however, that this is a reality of casework and that all laboratories are actively addressing this challenge.
Adelman: I could point to many labs being under-funded and backlogged, and I think the challenge of working in that reality is enormous, but my biased answer is a bit more personal: I think too many practicing forensic scientists aren’t sufficiently trained in statistical methods and mathematical literacy. If few forensic scientists understand the underlying mathematics of any given technical solution to a forensics problem, few of us are then able to offer constructive criticism and the field as a whole is somewhat lessened as a result. I think there is a fear of mathematics in some cases – it’s certainly something I’ve felt at various points in my career – and many of us are taught from the time we’re young that it’s okay to accept this fear as an end state.
What do you think are likely to be the most exciting developments for the industry over the next couple of years?
Marciano: The move to DNA sequence analysis is exciting. Enhancing the ability to interpret DNA mixtures allows for the interpretation of samples that may have been too complex for fragment analyses. This also will give birth to a new world of software and predicative tools that will permit new methods of genetic identification.
Adelman: I have to echo Mike’s answer here. Research involving sequence data is certainly an area of personal excitement.
What tips would you give to someone who is just starting out in the forensics field, or what’s the best advise you’ve ever received?
Marciano: Always remember that fundamental science underlies the methods and practices used in forensic science. Keep as current as possible and never forget that you are an applied scientist.
Adelman: The best advice I’ve ever received – and I think this broadly applies to all the sciences – is to repeatedly expose yourself to other ways of learning, thinking, and doing. Many important problems are already solved in a different discipline, just waiting to be discovered and implemented.
Do you have a hidden talent?
Marciano: I played football in college, offensive line, and I have a photographic memory.
Adelman: I once told a (debatably coherent) bedtime story to my eldest daughter that lasted over 100 nights. It remains to be seen whether I can duplicate that feat for the other two.
How did you become interested in forensics?
Marciano: About twenty years ago, as a senior in high school, I approached the local DNA technical leader, Kathy Corrado, and I asked her what it takes to be a forensic DNA analyst. I followed her advice; six years later she was the lab director who first hired me.
Adelman: My interest in forensics began in 2013 at the genesis of this project, though it has since expanded well beyond the boundaries of DNA mixture interpretation
For those who are on the fence about registering for the upcoming ISHI, please share your thoughts and reasons why they should attend.
Marciano: It is my personal favorite. ISHI is one of the few international conferences that centers on bringing the forensic DNA community together. In my opinion, it never disappoints; presentations include state-of-the-art science, updates on regulatory aspects of the field and current applied problem sets, such as backlog reduction. The conference itself is incredibly well planned and executed.
Adelman: This will be my first time attending. I’ve heard only excellent things, which perhaps has set an unfairly high bar!
WOULD YOU LIKE TO SEE MORE ARTICLES LIKE THIS? SUBSCRIBE TO THE ISHI BLOG BELOW!