Multiple sclerosis is an immune-mediated demyelinating disorder of the central nervous system. Because of the complexity of etiology, pathology, clinical manifestations, and the diversity of classification, the diagnosis of MS is very difficult. We found that McDonald Criteria is very strict and relies heavily on the evidence for DIS and DIT. Therefore, we hope to find a new method to supplement the evidence and improve the accuracy of MS diagnosis.
We finally selected GSE61240, GSE18781, and GSE185047 based on the GPL570 platform to build a diagnosis model. We initially selected 54 MS susceptibility locus genes identified by IMSGC and WTCCC2 as predictors for the model. After Random Forests and other series of screening, the logistic regression model was established with 4 genes as the final predictors. In external validation, the model showed high accuracy with an AUC of 0.96 and an accuracy of 86.30%. Finally, we established a nomogram and an online prediction tool to better display the diagnosis model.
The diagnosis model based on microarray data in this study has a high degree of discrimination and calibration in the validation set, which is helpful for diagnosis in the absence of evidence for DIS and DIT. Only one SLE case was misdiagnosed as MS, indicating that the model has a high specificity (93.93%), which is useful for differential diagnosis. The significance of the study lies in proving that it is feasible to identify MS by peripheral blood RNA, and the further application of the model and be used as a supplement to McDonald Criteria still need to be trained with larger sample size.