Biomarkers and molecular endotypes of sarcoidosis: lessons from omics and non-omics studies

Abstract

Sarcoidosis is a chronic granulomatous disorder characterized by unknown etiology, undetermined mechanisms, and non-specific therapies except TNF blockade. To improve our understanding of the pathogenicity and to predict the outcomes of the disease, the identification of new biomarkers and molecular endotypes is sorely needed. In this study, we systematically evaluate the biomarkers identified through Omics and non-Omics approaches in sarcoidosis. Most of the currently documented biomarkers for sarcoidosis are mainly identified through conventional "one-for-all" non-Omics targeted studies. Although the application of machine learning algorithms to identify biomarkers and endotypes from unbiased comprehensive Omics studies is still in its infancy, a series of biomarkers, overwhelmingly for diagnosis to differentiate sarcoidosis from healthy controls have been reported. In view of the fact that current biomarker profiles in sarcoidosis are scarce, fragmented and mostly not validated, there is an urgent need to identify novel sarcoidosis biomarkers and molecular endotypes using more advanced Omics approaches to facilitate disease diagnosis and prognosis, resolve disease heterogeneity, and facilitate personalized medicine.

Keywords: biomarker; endotype; machine learning algorithms; omics; sarcoidosis.

Figure 1

General workflow of Omics study to identify biomarkers and endotypes. First, liquid specimens and tissue samples collected from exploratory cohort are prepared ready for the analysis on the omics platforms. Second, omics datasets, including genomics, transcriptomics, proteomics, metabolomics, and microbiomics, are generated by experiments. Next, the datasets are pre-processed before statistical and machine learning analyses. Statistical analyses include differentiation expression tests, correlation analysis, logistic regression, Bayesian model, etc. Machine learning algorithms encompass supervised methods (e.g., gradient boosting, deep neural network, and laten classification and prediction) and unsupervised methods (e.g., PCA, k-means clustering, hierarchical clustering). Furthermore, network analysis can be performed with weighted gene co-expression network analysis (WGCNA), weighted protein correlation network analysis (WPCNA), genome-wide association study (GWAS) and meta-analysis. In general, function annotations are essential to rank differential gene-, transcript-, protein-, or metabolite-related signaling pathways, functions, and interactions. These analyses help identify candidate biomarkers, molecular endotypes, and unique phenotypes, along with developed predictive models. Specific metrics such as ROC AUC c statistic, odds ratio, risk ratio, and others are applied for biomarker comparisons. Finally, it is critical to validate the results in independent cohorts of multi-ethnic origins and under-represented populations, particularly since sarcoidosis is common in minorities.