Subtypes of Relapsing-Remitting Multiple Sclerosis Identified by Network Analysis
Quentin Howlett-Prieto1, Chelsea Oommen1, Michael Carrithers1, Daniel Hier2
1Department of Neurology and Rehabilitation, University of Illinois at Chicago, Chicago IL USA, 2Department of Electrical and Computer Engineering, Missouri University of Science and Technology, Rolla MO USA
Objective:
To use network analysis to identify subtypes of relapsing-remitting multiple sclerosis patients based on their cumulative signs and symptoms.
Background:
Network analysis can elucidate the complex relationships between the phenotypes, genes, proteins, and metabolic pathways that underlie human diseases, without a priori knowledge of how these elements will interact. It is of interest to determine whether subtypes of multiple sclerosis based on predominant clinical presentation (sensory, cerebellar, motor, visual, cognitive, fatigue, etc.) can be identified. Clinical subtypes could differ regarding prognosis, course, and response to treatment.
Design/Methods:
We reviewed the electronic medical records of 120 patients with relapsing-remitting multiple sclerosis and recorded signs and symptoms. Signs and symptoms were mapped to a neuro-ontology and then collapsed into 16 superclasses by subsumption. Bipartite (subject-feature) and unipartite (subject-subject) network graphs were created using Gephi for both a normalized and unnormalized feature set. Degree and weighted degree were calculated for each node. Graphs were partitioned into communities using the modularity score. Feature maps were used to visualize differences in features by community.
Results:
Network analysis of unipartite graphs yielded higher modularity scores (unnormalized 0.377, normalized 0.49) than the bipartite graphs (unnormalized 0.258, normalized 0.247). Across all graphs, we found large communities of subjects with predominantly sensory features and communities with predominantly motor features (weakness combined with hypertonia, hyperreflexia, or incoordination). When the feature dataset was normalized, network analysis showed motor and sensory communities and additional communities of subjects with a predominance of fatigue, pain, and cognitive features.
Conclusions:
Network analysis can partition multiple sclerosis subjects into communities based on signs and symptoms. Communities of subjects with predominant motor, sensory, pain, fatigue, cognitive, and behavior features were found. Larger datasets and additional partitioning algorithms are needed to confirm these results and elucidate their clinical significance.