Digital Forensics and Sensory Forecasting through VOC Analysis

Everyone leaves a trace, whether it's a tangible object, invisible DNA, or even an odor. 

In a recent study, a team of scientists achieved a remarkable 96% accuracy in determining human sex using a machine learning model guided by human expertise. Researchers collected hand odor samples from 60 individuals and analyzed them using Headspace-Solid Phase Microextraction-Gas Chromatography-Mass Spectrometry (HS-SPME-GC-MS). The results revealed distinct VOC signatures that allowed for the classification and prediction of gender. Various dimensional reduction techniques were employed to interpret the data, such as Partial Least Squares-Discriminant Analysis (PLS-DA), Orthogonal-Projections to Latent Structures Discriminant Analysis (OPLS-DA), and Linear Discriminant Analysis (LDA). The highest discrimination and classification of subject gender were observed with OPLS-DA and LDA as confidence level ellipses of both models were not seen to intersect. 

In another study, a combination of deep learning, chemometrics, and sensory evaluation proved effective in distinguishing between various methods of roasting food. The researchers employed E-nose and E-tongue devices, quantitative descriptive analysis (QDA), HS-GC-IMS, and HS-SPME-GC–MS to differentiate lamb shashliks prepared through traditional charcoal grilling and four alternative methods. The results showed that these techniques effectively identified the characteristic flavors and volatile organic compounds (VOCs) associated with each roasting method. The clustering heat maps were generated using TBtools and Python was used to run SVM, RF, XGBoost, DNN 5-layer, CNN-SVM, and t-SNE. The CNN-SVM model outperformed other models in predicting VOC content and identifying the specific roasting methods. 

REFERENCES

Chantrell J. G. Frazier ,Vidia A. Gokool ,Howard K. Holness,DeEtta K. Mills,Kenneth G. Furton. Multivariate regression modelling for gender prediction using volatile organic compounds from hand odor profiles via HS-SPME-GC-MS Published: July 5, 2023

https://doi.org/10.1371/journal.pone.0286452

Shen C, Cai Y, Ding M, Wu X, Cai G, Wang B, Gai S, Liu D. Predicting VOCs content and roasting methods of lamb shashliks using deep learning combined with chemometrics and sensory evaluation. Food Chem X. 2023 Jun 14;19:100755. doi: 10.1016/j.fochx.2023.100755. PMID: 37389322; PMCID: PMC10300318.

Of blood and breath: metabolite-based diagnosis of ovarian cancer

Physicians always knew that breath contains clues to diseases. Chemicals in breath often correlate with chemicals in saliva and blood - be it alcohol, anaesthetics or other metabolites (see, for example, this study by Dr Andreas Hengstenberg).

As one of my interests is breath-based detection of ovarian cancer, I took note of the recent paper claiming 99% to 100% accuracy of detecting ovarian cancer by metabolites in blood.
The authors used customized functional support vector machine-based machine-learning algorithms to classify thousands of metabolites measured by mass spectrometry (JEOL AccuTOF™ DART® that allowed to forego conventional liquid chromatography as sufficient resolution was achieved without separation) in peripheral blood. 

100% sensitivity and 100% specificity was achieved with 64-30 split validation technique, while 100% sensitivity and 98% specificity was the accuracy of leave-one-out-cross-validation. Very large number of metabolites, from 2,000 to 3,000 features, contributed to such discriminatory power (see the list of 14,000+ in supplemental material) 
 

Set of 25 canonical metabolic pathways relevant to the uploaded elemental
formulae ranked according to their p-values (hypergeometric distribution).

Histamine, amino acid, fructose and glucose metabolism were among the most prominent processes discriminating cancer and healthy blood.
It's that simple: sugar feeds cancer. Scientists have long found that cancer cells slurp fructose, and that fructose intake can be linked to some cancers. Histamine/polyamine interplay in cancers is also known. Histamine may be involved in inhibition of the local immune response against cancer. Is amino acid metabolism also linked to cancer? Well, what is not.   




Metabolomic biomarkers were always known to have diagnostic potential - cholesterol and glucose are among the oldest and most widely performed diagnostic tests. Yet, most bleeding edge cancer detection platforms are genomic or proteomic in nature.  Of the thousands of known biomarkers, only a handful have made it into the clinic. Existing ovarian cancer tests mostly rely on detecting a protein -  carbohydrate antigen 125. Vermillion's OVA1 and HealthLinx OvPlex tests use five proteins. This may be extended to 7.

Metabolites represent the end products of the genome and proteome, thus metabolomics-based diagnostics  holds the promise of providing powerful diagnostics,  allowing for differentiation of increased and decreased levels of chemicals with low process coefficient of variation.


Metabolomic tests were used for medical diagnostics starting with Hippocrates and Lavoisier. They continue to be explored by modern scientists. Dr Michael Phillips, for example, developed HeartsBreath Test, approved by the US Food and Drug Agency for early diagnosis of heart transplant rejection. Research proved the potential of inexpensive breath tests in discriminating lung, breast, colon and prostate cancers. Let's hope the new article  - along with others - will lead to novel consumer products, not only more academic research and peer-reviewed publications.



Zhou M, Guan W, Walker LD, Mezencev R, Benigno BB, Gray A, Fernández FM, & McDonald JF (2010). Rapid Mass Spectrometric Metabolic Profiling of Blood Sera Detects Ovarian Cancer with High Accuracy. Cancer epidemiology, biomarkers & prevention : a publication of the American Association for Cancer Research, cosponsored by the American Society of Preventive Oncology PMID: 20699376