<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=648880075207035&amp;ev=PageView&amp;noscript=1">

Presentation on September 28-29, 2017, at the 5th Annual BayesiaLab Conference:

Evaluating the link between microbiome and cosmetic clinical signs with PLS and Bayesian approaches

Marie Thomas, L’Oréal Research and Innovation, Aulnay-sous-Bois, France, mthomas@rd.loreal.com

Alban Ott, Life&Soft,  Fontenay-le-Fleury, France, aott@lifeandsoft.com 


Skin microbiome is a new emergent biologic target in cosmetic. Gut microbiome is well known and already linked to some diseases. What about the skin? Identifying changes in biodiversity and micro-organisms on the skin surface that are associated with a specific state might be a new opportunity to explain skin disorders and prevent or correct them. That’s easier said than done. Microbiome analyses lead to specific data: basically counting of huge number of organisms, sparse and correlated. Some of these counts are very low and seem quite irrelevant on a biological point of view. This makes it difficult to determine the link between hundreds or even thousands micro-organisms, with often very slow dynamic, and clinical signs of interest. In order to address this problem, with low to inexistent state of the art, we used several methods: First to select microorganisms of interest, we used variable selections such as sparse ones. We also applied a specific “omics” method: DESeq2 which is a differential gene expression analysis based on the negative binomial distribution. We also did it manually based on the top abundances and with descriptive statistics such as effect-sizes. Then, to discriminate skin types and correlate with skin disorders, we used Partial Least Squares approaches (PLS Discriminant Analysis), and Bayesian Networks. We will present and compare results of a concrete successful example: “Recovery of the skin after a harsh cleansing on sensitive skin”, focusing on disorders like itching or dryness of the skin (linked to sensitive skin). Thanks to PLS DA, We are able to predict the skin type thanks to microbiome, clinical and instrumental measurements. We also identified some hits that are correlated with several clinical signs and instrumental measurement. Those results are confirmed in the second Bayesian approach that gave very similar results and allowed a high level of visualization and interpretation. Today we need the feedback of microbiologists to validate the biological relevance of the hits.

Keywords: Microbiome, DESeq2, Partial Least Squares, variable selection, Bayesian Network.

Presenter Biography

Marie THOMAS.jpgAfter an academic background (MBA of methodology and statistics for biomedical research), and several years spent in pharmaceutical domain, Marie Thomas had joined the L’OREAL’s research and innovation division in 2003. The primary aim of this division is to deliver new innovative products. To answer this question, lots of knowledge studies are conducted to better understand the link between biology and clinical skin disorders. Microbiome is a new domain and needs to be investigated but promise to deliver new markers and target. Marie Thomas is in charge of statistical analyses related to this specific area.