Subtyping insomnia disorder with a population graph attention autoencoder: revealing two distinct biotypes

Area of Science:

• Neuroimaging
• Computational Psychiatry
• Sleep Medicine

Background:

• Insomnia disorder (ID) is neurobiologically diverse, making traditional group-level neuroimaging insufficient for characterization.
• Subtyping ID using neuroimaging and clinical data can reveal biologically and clinically relevant subgroups.

Purpose of the Study:

• To develop and apply a Gray Matter Population Graph Attention Autoencoder (GM-PGAAE) for identifying distinct subtypes of insomnia disorder.
• To integrate structural magnetic resonance imaging (MRI) and clinical data for a comprehensive understanding of ID heterogeneity.

Main Methods:

• Developed GM-PGAAE integrating atlas-based gray matter volumes and clinical similarity to create an adjacency matrix.
• Employed a Graph Attention Autoencoder to learn low-dimensional embeddings and cluster them to identify subtypes.
• Utilized Voxel-Based Morphometry (VBM) and individualized differential structural covariance networks (IDSCNs) for regional and network-level analyses.

Main Results:

• Identified two distinct subtypes of insomnia disorder.
• Subtype 1 showed higher symptom severity and significant gray matter reductions in specific brain regions (cerebellar vermis, thalamus, middle occipital cortex, fusiform gyrus, paracentral lobule) compared to Subtype 2.
• Subtype 1 exhibited negative associations between gray matter volume and clinical scores, with reduced thalamocortical and subcortical Z-scores in IDSCNs.

Conclusions:

• The GM-PGAAE framework successfully integrates structural MRI and clinical data to delineate biologically distinct subtypes of insomnia disorder.
• The identified subtypes represent specific neurobiological profiles within the broader insomnia disorder diagnosis.
• These findings pave the way for more personalized diagnostic and treatment approaches for insomnia.