Discriminative Machine Learning Analysis for Skin Microbiome: Observing Biomarkers in Patients with Seborrheic Dermatitis

Research Articie September 06,2022

H.E.C. van der Wall ¹ , R.J. Doll ² , G.J.P. van Westen ³ , T. Niemeyer-van der Kolk ⁴ , G. Feiss ⁵ , H. Pinckaers ⁶ , M.B.A. van Doorn ⁷ , T. Nijsten ⁸ , M.G.H. Sanders ⁹ , A.F. Cohen ¹⁰ , J. Burggraaf ¹¹ , R. Rissmann ¹² , L.M. Pardo ¹³ hide author's information

1 : Method Development, Centre for Human Drug Research, The Netherlands
2 : Method Development, Centre for Human Drug Research, The Netherlands
3 : Drug Discovery and Safety, Leiden Academic Centre for Drug Research, The Netherlands
4 : Method Development, Centre for Human Drug Research, The Netherlands
5 : GL Feiss Consulting, LLC, Telford, Pennsylvania, United States
6 : Department of Pathology, Radboud University Medical Center, The Netherlands
7 : Department of Dermatology, Erasmus Medical Center, The Netherlands
8 : Department of Dermatology, Erasmus Medical Center, The Netherlands
9 : Department of Dermatology, Erasmus Medical Center, The Netherlands
10 : Method Development, Centre for Human Drug Research, The Netherlands
11 : Method Development, Centre for Human Drug Research, The Netherlands
12 : Method Development, Centre for Human Drug Research, The Netherlands
13 : Department of Dermatology, Erasmus Medical Center, The Netherlands

DOI: 10.55578/joaims.220819.001

Keywords: Microbiome, Artificial intelligence, Machine learning, Seborrheic dermatitis, Data science

Full Text

PDF

Download Citation

Abstract

In recent years the skin microbiome has taken center stage as drug target and as disease biomarker. Computational analyses of microbiome sequencing data from patients with skin diseases, for example seborrheic dermatitis, can be performed to identify discriminative biomarkers in the microbiome profile. The aim of the present study was twofold, namely to employ machine learning to predict disease from the microbiome dataset, and to identify discriminative biomarkers in the microbiome of patients with seborrheic dermatitis versus healthy controls using machine learning techniques. The population consisted of 97 patients with seborrheic dermatitis and 763 healthy controls. Skin swabs were taken from naso-labial fold (lesional skin: n = 22; non-lesional skin: n = 75, controls: n = 763). Using an extra trees machine learning model, differences between the skin microbiome of patients with seborrheic dermatitis versus healthy controls were characterized. Subsequently, the most important microorganisms for discrimination were determined by feature analysis and SHapley Additive exPlanations (SHAP) values...