Category: Autoimmune rheumatologic diseases
Currently, diagnosis and monitoring the disease progression of Rheumatoid Arthritis (RA) is challenging and there is no a biochemical test for detection early-stage RA. In this study, we aimed to identify putative biomarkers for RA by leveraging gene expression data with machine learning approaches.
We collected publicly available microarray data with 348 synovium (SY) and 2,518 whole blood (WB) samples. The raw data was processed, merged and normalized across studies and treatments. We developed a machine learning pipeline for robust feature selection. We found 16 genes highly associated with RA in both tissues: 6 up-regulated: MARCH1, NMI, LXN, DCP2, IFT20, RPS27L, and 10 down-regulated: DEXI, GPATCH8, BRD2, BMS1, ANKRD11, CBLB, TNPO2, MYC, ZBTB16, PRDX6. The up-regulated genes are involved in immune and inflammatory response, apoptotic processes, and DNA damage response, whereas the down-regulated genes are involved in in tissue homeostasis and bone development, negative regulation of epidermal growth factor-activated receptor activity, and response to oxidative stress, regulation of cell growth and proliferation, oxygen transportation, and regulation of gene expression. Finally, we built a prediction model with selected genes on the WB tissue, obtaining AUC 0.97 with sensitivity 0.85 and specificity 0.95. The validation of results was conducted on an independent synovium RNA-seq data with the model performance of AUC 0.97 with sens. 0.71 and spec. 0.99.
In our comprehensive in-silico search we found novel biomarkers in RA. Identification of extensive proteins secretion in blood could allow precision phenotyping which could have a positive impact on monitoring disease progression and patient treatment.