Proteomic Approaches to Defining Remission and the Risk of Relapse in Rheumatoid Arthritis

Area of Science:

• Rheumatology
• Proteomics
• Machine Learning

Background:

• Rheumatoid Arthritis (RA) patients achieve remission, but mechanisms of disease persistence and flare risk are unclear.
• Understanding stable RA heterogeneity is crucial for managing long-term outcomes and treatment withdrawal.

Purpose of the Study:

• Identify serum proteomic alterations associated with stable RA features.
• Develop a machine learning model to predict future RA flare using baseline proteomics.

Main Methods:

• Analyzed 1307 serum proteins from 130 stable RA patients in remission (DAS28<2.6) using SOMAscan.
• Applied unsupervised clustering and supervised classification (XGboost) to identify proteomic biomarkers and predict flare after 12 months.
• Utilized network analysis to explore enriched pathways in proteomic datasets.

Main Results:

• Identified 4 proteomic clusters; one cluster showed lower DAS28 scores linked to humoral immunity and complement activation.
• Proteomic clustering did not reliably predict flare risk.
• An XGboost model accurately predicted relapse (AUC=0.80) using baseline serum proteomics.

Conclusions:

• The serum proteome offers insights into stable RA and its clinical variations.
• Combining proteomics with machine learning shows promise for predicting RA flare in patients attempting therapy withdrawal.