Most age estimation models in forensic DNA were built on a linear assumption: that methylation changes at a predictable, constant rate across the lifespan. Maria Flores, a PhD candidate at UCLA, didn’t think that was right.
Her research develops a non-linear model for age estimation using DNA methylation data — an approach she argues better reflects how markers like ELOVL2 actually behave across age groups. The results from her first validation dataset were encouraging: her model reduced mean error from 4.6 years (as reported in the original paper) to 3.73 years.
Flores presented her poster at ISHI 36 as a Student Ambassador. We sat down with her to learn more about the work, what drove her to this research, and what she hopes the forensic community takes away from it.
What sparked this research, and what have you found so far?
Flores entered this area through a lab rotation with Dr. Matteo Pellegrini’s group at UCLA, where DNA methylation research was already underway. What stood out to her was a gap in how the field approached age estimation.
“I knew DNA methylation was being applied to age estimation models in forensics, but I realized a key limitation: most existing approaches relied on linear regression. This felt counterintuitive since age-associated methylation markers, such as ELOVL2, show non-linear patterns over the lifespan.”
Using two publicly available datasets, she trained and validated a non-linear model. The first dataset produced a clear improvement over the published benchmark. The second was more complicated — her model didn’t beat the original error rate overall, but it did reduce systematic bias using residual modeling.
“We were able to actually get a better estimate in age than the original data set from the original paper that I used. The original paper had a mean error of 4.6 years, and we were able to get that down to 3.73 years with our model, which is very exciting.”
Why does this work matter beyond the accuracy numbers?
For Flores, improving prediction accuracy is inseparable from the question of who those predictions work for. She’s explicit that forensic science has not always been applied equitably — and that designing for equity has to be built into the research process, not added at the end.
“Scientific tools have not always been applied equitably across all groups, making it especially important to me to approach this work in a way that ensures it is beneficial for all communities.”
Her next direction reflects this: she wants to expand validation to larger, more diverse datasets, so that any model she builds can be tested across the broadest possible range of individuals before it reaches practice.
What’s one part of the process that was unexpectedly challenging?
Flores was candid about the friction that comes with computational research — specifically the hunt for usable, publicly available datasets and the slow process of debugging code when something isn’t working.
“Computation can be discouraging at times, especially when a bug or small mistake sends you back through your code to trace the source.”
What kept her going was the understanding that research progress happens incrementally, and that patience with the process is as important as technical skill.
“I did find it exciting that after all that frustration our approach outperformed the results reported in the original study from which one of the datasets was derived.”
What do you want people to walk away from your poster knowing?
Her answer wasn’t just about findings. It was about communication.
“It is important to me that I can explain my work in a way that is accessible and easy to follow.”
That orientation — clarity over complexity — runs through how she talks about her research more broadly. The goal isn’t to impress; it’s to be useful to the people working cases.
Where is this research headed next?
Flores is looking past age estimation to the broader landscape of what DNA methylation can tell us. She’s interested in how epigenetic markers might surface additional information useful to forensic investigations — not just age, but other biological signals that could strengthen identification work.
“I want to continue pursuing research questions that ensure forensic tools are fair and considerate of all groups.”
She’s also thinking through her own career path, weighing forensic consulting against time in a crime lab to build hands-on bench experience.
“I am strictly computational and don’t really have any experience in the lab or bench. So getting that experience would be amazing as well and seeing where that will catapult me.”
For a field that’s still developing its toolkit, that combination — computational depth, equity-minded design, and a willingness to go where the work takes you — is exactly what the next generation of forensic science looks like.