Abstract
Purpose :
Current analysis of cell populations in body fluids from patients with ocular diseases relies strongly on cytometry, which measures the expression of markers on each cell. A recent study1 combined multiparameter single cell analysis with machine learning classification to accurately predict patients with Behcet’s Disease (BD) and patients with sarcoidosis on the basis of five markers on CD8+ cells. We have now extended the numbers of patients analysed and incorporated patients with other ocular diseases.
Methods :
Peripheral blood mononuclear cells (PBMC) was isolated from patients with BD (n=-100), sarcoidosis (n=15) isolated idiopathic uveitis (n=15) and birdshot uveitis (BU; n=15) and healthy controls (n=45). PBMC were labelled with a 15-colour antibody panel and the data was collected using flow cytometry and subsequently compensated using FlowJo. Compensated data was then analysed by two machine learning algorithms, Supercell, which randomly allocates multiple single cells into a supercell and calculates a single score value for all parameters which are then compared between patient groups to identify differences; and quantile-based analysis which compares each parameter against all others to identify the most significant phenotype which can discriminate between patient groups.
Results :
The results show that all disease groups can be distinguished from healthy controls via supercell and quantile-based analysis. In patients with BD this was based on markers including CD4 (IL22, p=0.00015) CD8 (TNF-α and IL-23R p=1x10-5) supporting previous findings by protein and genomic studies. Patients with ocular BD could be distinguished from patients without eye involvement by markers such as TNF-α, IL23R and IL17 (p=0.001). Between diseases patients with BD could be distinguished from patients with Birdshot uveitis CD8 (IL22 and CD27 p=1x10-9).
Conclusions :
Flow cytometry has been a hugely influential technique in advancing our understanding of the cellular basis of ocular disease. Novel machine learning algorithms increase the range of analysis to distinguish between diseases with a similar aetiology. The ability to apply such techniques to include other parameters such as gender, genetics and therapy have exciting potential.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.