Estimating a brain network predictive of stress and genotype with supervised autoencoders

Area of Science:

• Neuroscience
• Computational Psychiatry
• Systems Biology

Background:

• Mental illnesses present complex challenges for targeted treatment.
• Current brain stimulation protocols often lack precise, individualized targeting.
• Understanding multi-region brain electrical dynamics is crucial for effective intervention.

Purpose of the Study:

• To develop a computational approach for designing targeted brain stimulation protocols.
• To identify relevant multi-region electrical dynamics associated with specific mental health outcomes.
• To leverage supervised autoencoders (SAEs) for improved predictive modeling of brain network activity.

Main Methods:

• Modeling multi-region electrical dynamics as a superposition of latent networks.
• Utilizing supervised autoencoders (SAEs) to enhance predictive performance.
• Implementing modeling constraints to ensure biological relevance and interpretability.

Main Results:

• Identified a specific latent network associated with stress responses.
• Validated the network's relevance by aligning it with a previously established stimulation protocol.
• Characterized a genotype associated with bipolar disorder through network analysis.

Conclusions:

• The developed approach effectively models complex brain dynamics for stimulation protocol design.
• SAEs offer improved predictive power in modeling neural activity for clinical applications.
• This framework provides a biologically relevant method for investigating neural correlates of mental disorders and guiding therapeutic interventions.