A Clinically and Biologically Based Subclassification of the Idiopathic Inflammatory Myopathies Using Machine...

Area of Science:

• Rheumatology
• Computational Biology
• Immunology

Background:

• Idiopathic inflammatory myopathies (IIMs) exhibit significant heterogeneity, limiting the accuracy of existing predictive models.
• Computational approaches offer a potential solution by stratifying patients based on clinical and biological characteristics.

Purpose of the Study:

• To apply unsupervised machine learning, specifically similarity network fusion (SNF), to identify novel patient subgroups within IIMs.
• To develop a predictive classifier for these newly identified patient groups.

Main Methods:

• Similarity Network Fusion (SNF) was applied to clinical and biological data from 168 IIM patients (adult polymyositis, adult dermatomyositis, juvenile dermatomyositis).
• A sparse multinomial regression model was developed to classify patients into identified groups.
• Chi-squared tests were used to associate patient groups with established myositis subtypes.

Main Results:

• SNF identified five distinct patient groups that further subdivided existing myositis subtypes.
• Autoantibody profiles (anti-Mi-2, anti-SRP, anti-NXP2, anti-synthetase) defined four groups, while immunoglobulin M (IgM) depletion defined the fifth.
• Specific associations were found between groups and subtypes: adult DM with anti-Mi-2/anti-synthetase, JDM with anti-NXP2, and adult PM with IgM depletion/anti-SRP.

Conclusions:

• Unsupervised machine learning successfully identified clinically and biologically homogeneous patient groups within IIMs.
• These findings provide a foundation for an integrated disease classification system that incorporates both clinical and biological phenotypes.
• The identified groups offer a more refined understanding and classification of IIM subtypes.