Human Skin Microbiome: Selecting Informative SNPs for Human Identification

ISHI Ambassador, Allison Sherier of UNTHSC, describes her PhD thesis work, in which she hopes to use the human skin microbiome as an alternative source of DNA to identify evidence from crime scenes.





Travis: Good morning, Allison. How are we doing today?


Allison: I’m doing great. How are you?


Travis: I’m doing fantastic, thanks. Hey, I was wondering, do you think you could share with us the PhD thesis that you brought with you this year?


Allison: Yes, I would love to. So, my research is over the human skin microbiome as an alternative source of DNA to identify evidence from crime scenes. So, obviously, we’re not quite to the collecting evidence from crime scenes part of it, but we really hope that this will be a good new marker system for human identification.


So, my project is focusing on skin samples that were taken from individuals, and looking at whether we can accurately predict which individual those skin samples came from.


So we took three replicate swabs on either the hand, the foot, or the manubrium, which is up here on the collar bone, and then, previous work had shown our classification accuracies for those samples using particular algorithms was about 70-80%. Hopefully, I’m going to get that a lot closer to 100%.


So, initially, I was looking at samples from the same individual. So, we consider the microbiome on them as a population. So, think about it as the populations that we use in forensics. We’re talking about the same sort of thing, but now we’re talking about microbes on a single person.

We’re looking at replicate one versus replicate two versus another replicate on the same individual and seeing that those two are consistent.


Then, we went on to comparing what I’m calling incorrect or non-associated individuals. So, these are two different individuals that are being compared against each other. So, if it’s incorrect, that means in a previous study, we showed that maybe individual one classified as individual five. Well, obviously that’s not true. The sample came from individual one, which we know, so I’m trying to look to see if my method better identifies individual one as individual one instead of whoever they were misclassified as.

The non-associated is an individual that was correctly identified with all previous algorithms that were previously tried with these samples. And, so far, looking at FST, which most of our population geneticists here are familiar with, we’re starting to see pretty good trends for FST values. So, our red lines here are samples from the same individuals, so we have low FST, which means their populations look very similar. And then, in our non-associated or incorrect individuals, we start seeing higher FST values going out here to the right, which means their populations do not look similar to each other, which is exactly what we want to see.


So, so far, all good. It looks like my PhD thesis is working, so that’s always very exciting to not have to go back to the drawing board. And then, hopefully next year, I’ll be back with a lot more information for our poster and maybe a better marker system too.


Travis: That’s fantastic. So, what you’re doing here, is this all a computer model and where did you get your information from that you used for this?


Allison: So, the samples are real samples. They were collected by another PhD student that worked on this project before me, and she worked on the marker design for our MPS marker system (we call it the HIDSkinPlex). But, now, for me, it is all just data analysis. So, I luckily had other students that had to do all the benchwork side of this and now just have piles and piles and piles of files to work with, so it’s been a great experience.

Travis: Well that’s fantastic. Thank you so much for taking time to present for us. We’ll see you later, thanks!