The skin microbiome holds the secrets to a whole host of health, wellbeing and personal care related questions: Why is my skin dry? Why do some people get acne and not others? Why does what I eat affect how my skin looks? The microbiome is also shaped by our lifestyle choices – it arguably could provide the clues to trace our day-to-day activity. Now, with the application of artificial intelligence (AI), there are promising developments that will enable us to decipher an individual’s skin microbiome and what this means for their health and skin care.
While there has already been a rise in products that target our skin microbiome by incorporating prebiotic and probiotic ingredients (like these), there is still a limited understanding of the complex interactions between each individual’s microbial composition and our physical characteristics. Well, maybe not for much longer…
Applying AI to skin microbiome research
The fusion of sophisticated analytical methods – such as AI and machine-learning – with skin microbiome data offers the potential to accelerate the development of novel, personalized and more effective products. Enter ‘Explainable artificial intelligence’ or EAI.
Let us explain (pun intended) what this means. EAI is an approach for the analysis of the skin microbiome and its relationship with aspects of human wellbeing. The ‘explainability’ aspect of the approach aims to provide clear reasons to justify the predictions made by the technology, which in this case are explained by the variations in microbiome composition. This aspect is a critical part of the researchers’ application of AI as it allows them to validate the predictive power of the model, as well as the claims underlying any recommendations made based on the insights gained.
Putting this into context, a team of researchers used leg swab samples from almost 1,500 women from Canada and the UK and developed an EAI model capable of identifying key types of bacteria present on human skin that can be used to predict the physical traits of an individual. The traits focused on skin hydration, age, menopausal status and smoking status. By assessing the presence and relative abundance of these bacteria, the researchers could identify distinct ‘microbial signatures’ and link these to physical differences observed for each of the four traits.
So, how is this research useful?
The EAI model offers hope for using the intricate and complex ecosystems of microbes that reside on our skin to inform and tailor our treatment or skin care routines. It generates actionable insights for skin care and health products. Effective translation of the data and learnings can accelerate development of more personalized products to support healthy skin, and reveal new information on the skin ageing process. Although the study focused on personalized care and general wellbeing, the approach can also be applied to predict the presence or status of other physical conditions using microbiome samples. This could help development of microbiome-based personalized therapeutics for disease, and non-invasive diagnostics and health monitoring.
Within this research, the model was able to identify several microbial members that were crucial for predicting differences in more than one of the four traits studied. This revealed key bacterial types that could prove useful for simultaneously monitoring (or even treating) a number of our physical characteristics if targeted in research. For example, high abundance of the bacterium Lactobacillus predicts higher skin hydration, younger age and pre-menopausal status (read more about Lactobacillus here). However, for accurate predictions of each trait, distinct combinations of bacterial members are needed. Higher abundance of Abiotrophia and Kocuria, for instance, drove the prediction of lower skin hydration, while Atopobium and Enterococcus together are most significant for age prediction.
These findings come to life when looking in detail at some of the traits studied.
Several types of bacteria predictive of skin hydration were revealed, with the identification of 11 bacteria associated with hydrated skin (Lactobacillus, Brevundimonas and Corynebacterium – all of which have been previously reported as linked to skin hydration or skin health[2–5]), and 19 bacteria associated with dehydrated skin (including Abiotrophia, Kocuria, Veillonella and Erwinia – also aligning with previous studies[6-8]). Of all the bacterial members identified, Lactobacillus was the most significant for the prediction of skin hydration level, followed by Corynebacterium.
The variations in microbial composition linked to skin hydration can provide insights to help predict the effect of skin care and hygiene treatments, supporting the development of personalized products that promote healthy skin – and a healthy skin microbiota.
Key microbes predictive of age were also identified, including 26 bacteria linked to younger age (such as Alloprevotella, Granulicatella, Gemella and Lactobacillus), and 4 bacteria linked to older age (Bacteroides, Pseudomonas, Bergeyella and Bacillus). Within this array of bacteria, Propionibacterium and Granulicatella were revealed to be key bacteria for predicting younger age. Bacillus, on the other hand, was the key bacterium driving accurate prediction of older age in the study, and has also previously been found to be dominant in the cheek microbiome of older people. Interestingly, some overlap between the bacteria identified as predictors for skin hydration and age was found. However, the majority (70%) of the bacteria that were most significant in predicting the status of each trait were different.
Previous research has suggested that the skin microbiome better predicts age than the microbiota of the gut and oral tracts, yielding predictions within four years of an individual’s actual age. The ability to infer key compositional changes in the skin microbiome as people grow older may also offer insights into the ageing process and help develop products that counteract skin ageing.
Using the EAI approach, the researchers demonstrated that the skin microbiome can accurately predict menopausal status, with key bacteria identified including Lactobacillus, Brevibacterium, Blastococcus, Parabacteroides and Streptococcus.
Lactobacillus was the most significant bacterium, predicting pre-menopausal status when high in abundance, whereas a high abundance of Streptococcus predicted post-menopausal status. This reflects a previous study, where Lactobacillus was observed to be the most abundant bacterium in the vaginal microbiome of pre-menopausal women (representing 64.4% of the microbiota), and its levels were found to be much lower post-menopause (24.4%) where it was replaced by Streptococcus (5.1%), among others.
Many of the bacteria considered to be important for this trait were also identified as key predictive bacteria for age. Lactobacillus, Granulicatella and Dermacoccus – for example – were among the most important bacteria in the prediction of younger ages, and they also predicted pre-menopausal status when present in higher numbers. Despite this crossover, the EAI model could accurately predict age separately within the pre- and post-menopausal groups, demonstrating that age prediction is robust despite possible confounding factors like menopausal status.
The power of predicting the onset of menopause through a simple scrub of the leg could be transformational to many women as a non-invasive approach, demonstrating the potential of this research to feed into the development of new and improved diagnostic and health or disease monitoring methods.
So, the microbiome landscape is complex. However, advanced technologies such as AI and machine learning can be applied to help researchers understand the intricate relationships between microbial communities and physical traits, and to translate data into actionable insights that support skin care, and our overall health and wellbeing.
In the Carrieri et al. study, a cohort of 62 Canadian women (21–65 years of age) displaying either healthy or moderately dry skin was used to collect a total of 1,200 leg skin microbiome samples which were processed via16S rRNA gene sequencing, as well as associated skin hydration characteristics (such as visual assessment of hydration level, pH, conductance and capacitance). Additional physical trait data (including age, menopausal status and smoking habit) were obtained from subject questionnaires. A second independent UK cohort of 278 samples from 102 women was also collected to investigate the generalizability of the EAI approach.
By combining the leg skin microbiome samples from the Canadian cohort with the EAI approach, the researchers could identify key bacteria associated with each of the physical traits, and subsequently use these to predict skin hydration, age, menopausal and smoking status for different subsets of the Canada cohort. The UK cohort served as a real-world test set, as it could be used to make and validate predictions using the machine-learning models trained on the Canada cohort.
For this EAI model, the explanations were expressed in terms of variations in the relative abundance of key microbes that drive the predictions.
1. Carrieri, A. P. et al. Explainable AI reveals changes in skin microbiome composition linked to phenotypic differences. Sci. Rep. 11, 4565 (2021).
2. Grice, E. A. & Segre, J. A. The skin microbiome. Nat. Rev. Microbiol. 9, 244–253 (2011).
3. Byrd, A. L., Belkaid, Y. & Segre, J. A. The human skin microbiome. Nat. Rev. Microbiol. 16, 143–155 (2018).
4. Kwaszewska, A., Sobiś-Glinkowska, M. & Szewczyk, E. M. Cohabitation—relationships of corynebacteria and staphylococci on human skin. Folia Microbiol. 59, 495–502 (2014).
5. Gao, Z., Tseng, C., Strober, B. E., Pei, Z. & Blaser, M. J. Substantial alterations of the cutaneous bacterial biota in psoriatic lesions. PLoS ONE 3, e2719 (2008).
6. Assarsson, M., Söderman, J., Dienus, O. & Seifert, O. Significant differences in the bacterial microbiome of the pharynx and skin in patients with psoriasis compared with healthy controls. Acta Derm. Venereol. 100, adv00273 (2020).
7. Zheng, H. et al. Altered gut microbiota composition associated with eczema in infants. PLoS ONE 11, e0166026 (2016).
8. Prod’homme, M. et al.Cutaneous infection and bactaeremia caused by Erwinia billingiae : a case report. New Microbes New Infect. 19, 134–136 (2017).
9. Kim, H.-J. et al. Segregation of age-related skin microbiome characteristics by functionality. Sci. Rep. 9, 16748 (2019).
10. Si, J., You, H. J., Yu, J., Sung, J. & Ko, G. Prevotella as a hub for vaginal microbiota under the influence of host genetics and their association with obesity. Cell Host Microbe 21, 97–105 (2017).