Machine learning is gradually revealing its potential in neuromuscular diseases. A Hispanic-American team recently used this artificial intelligence technology to characterize the gene expression profile specific to each type of inflammatory myopathy in muscle biopsies.
The study included 49 patients with autoimmune necrotizing myopathy, 39 with dermatomyositis, 8 with antisynthetase syndrome and 13 with sporadic inclusion myositis. Their muscle biopsies were compared to those of 20 control subjects. Different machine learning algorithms were trained on RNA sequencing data, with the results:
- the identification of a unique gene expression profile by type of myositis, but also by subtype of myositis depending on the auto-antibody produced by the patient: CAMK1G, EGR4 and CXCL8 are strongly expressed in the syndrome of antisynthetases, APOA4 is expressed only in autoimmune necrotizing myopathy with anti-HMGCR autoantibodies, MADCAM1 only in dermatomyositis with anti-Mi2 autoantibodies …;
- the good diagnostic performance (greater than 90% accuracy) of the algorithms for classifying muscle biopsies by type of myositis; the best performer succeeded in identifying the biopsies of patients with dermatomyositis in 92% of cases, even though histological analysis of the biopsy found evocative lesions (perifascicular atrophy) in only 72% of cases;
- the relevance of a differentiated therapeutic strategy because each type of myositis probably has a specific physiopathological mechanism, as evidenced by its unique gene profile; in dermatomyositis, for example, most genes strongly expressed in muscle are stimulated by interferon 1 (ISG15, MX1, etc.), confirming the central role of this mediator of the inflammation in this disease.
