Credit: FIU
Human odor is a complex mixture of volatile organic compounds (VOCs) secreted from the body, mostly attributed to the constant shedding of the epidermis of the skin. Dogs have been using these VOCs to reliably identify and track individual people for a very long time.
In the lab, however, it’s a different story. Laboratory-based identification of humans by odor profile has been difficult due to the lack of robust datasets and sufficiently developed analytical techniques, according to a team of forensic researchers at Florida International University.
In a new study, published in PLoS ONE, the team sought to further this research by demonstrating the efficacy of using VOC profiles for sex discrimination, comparison and classification. Although more research is needed, their analytical process/model resulted in a 96.67% accuracy rate for the identification of female and male subjects by hand odor.
Nearly 97% accuracy
For the study, the researchers collected samples from 60 volunteers—30 men and 30 women—of African American, Hispanic, and Caucasian race/ethnicity between the age of 18 to 46 years old. The participants were instructed to not wash their hands for 1 hour prior to sampling. The scientists used sterile gauze pads squeezed between the participants’ hands for 10 minutes to collect the sample.
First, the FIU team used headspace-solid phase microextraction-gas chromatography-mass spectrometry (HS-SPME-GC-MS) to analyze the hand odor samples. The subsequent VOC data was then evaluated with three different data models to test for accuracy—Partial Least Squares-Discriminant Analysis (PLS-DA), Orthogonal-Projections to Latent Structures Discriminant Analysis (OPLS-DA), and Linear Discriminant Analysis (LDA).
According to the study results, two of the models—PLS-DA and OPLS-DA—demonstrated clustering among the male and female subjects, revealing minimal discrimination between the sexes.
However, the LDA had an 96.67% accuracy rate for predicting and discriminating samples between male and female donors. One male sample and one female sample were misclassified using the model. The researchers say the misclassification of the two samples could be correlated to the abundance of a specific or group of compounds resembling a male subject profile and vice versa for a female subject.
“The overall impact of this study could assist in changing the trajectory of how we currently utilize human odor in the field of forensic. Investigators now have an opportunity to rapidly assess the volatiles of collected human odor samples,” the researchers write in their paper.
Time as a limitation
Of course, there is room for improvement. The FIU team says one of the biggest limitations of the study was the length of analysis time. The forensic researchers used an exhaustive method in terms of extraction and GC/MS run time in order to minimize the chances of overlooking informative compounds that could take longer to elute. Overall, this led to a lengthy—possibly undesirable—HS-SPME-GC-MS total analysis time. However, based on the findings of the study, the researchers say the method could be refined to allow for reduced collection analysis times.
Additionally, the sample size of the study was too small for the team to implement machine learning techniques or deep neural networks. However, natural language models like that could present a possibility if the sample data is deep enough.
Future implementation
The researchers say the overall statistical workflow they used in the current study could be applied to other identification factors, such as ethnicity/race and age when other discriminatory evidence may be lacking.
“Overall, the application of the models can be applied to various forensic data sets, and paves way for a tool that can perform standardized VOC comparisons,” said lead researcher Chantrell Frazier. “It could also be used in conjunction with detection canines. It can be one more tool for investigators to use to bring justice to victims of crime.”