rest2vec: Vectorizing the resting-state functional connectome using graph embedding.

Area of Science:

• Neuroscience
• Machine Learning
• Data Science

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for brain connectomics but faces analytical challenges, particularly in modeling negative correlations.
• Understanding the functional topology of the brain is complex, as functional relationships don't always align with anatomical distances.

Purpose of the Study:

• To develop a novel graph embedding pipeline, rest2vec, for analyzing functional connectomes derived from rs-fMRI data.
• To represent brain regions in a continuous 'functional space' using dimensionality reduction techniques.
• To identify biologically relevant connectivity gradients by analyzing 'functional distances' within this embedded space.

Main Methods:

• Utilized machine learning embedding techniques, inspired by word2vec, to create the rest2vec pipeline.
• Employed the phase angle spatial embedding (PhASE) method for dimensionality reduction of the functional connectome.
• Conceptualized rest2vec within the framework of maximum mean discrepancy (MMD) and kernel PCA.

Main Results:

• Successfully embedded the rs-fMRI connectome into a lower-dimensional vector space.
• Demonstrated that 'functional distance' in this space can uncover significant connectivity gradients.
• Mapped brain regions based on their functional relationships, offering a new perspective on brain organization.

Conclusions:

• rest2vec provides a powerful low-dimensional representation of the rs-fMRI connectome.
• The method enhances the understanding of the brain's functional organization by mapping regions according to their functional relationships.
• This approach offers a novel way to study brain connectivity and identify meaningful gradients.