Deep Learning in the Identification of Electroencephalogram Sources Associated with Sexual Orientation

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomarkers

Background:

• The biological underpinnings and potential neurofunctional correlates of sexual orientation remain largely unexplored.
• Advancements in deep learning enable sophisticated analysis of complex biological datasets, including electroencephalogram (EEG) data, without predefined feature selection.
• Investigating electrophysiological differences associated with sexual orientation could reveal novel biological insights.

Purpose of the Study:

• To apply deep learning algorithms to classify resting-state EEG data from males based on sexual orientation.
• To identify specific neurofunctional patterns that differentiate homosexual and heterosexual males using advanced analytical techniques.

Main Methods:

• Utilized three cohorts: homosexual men, heterosexual men, and a mixed-sex group.
• Employed a newly trained deep learning network for sexual orientation classification.
• Applied Gradient-weighted Class Activation Mapping (Grad-CAM) and source localization to identify distinguishing spatiotemporal EEG patterns.

Main Results:

• A deep learning model achieved 83% accuracy in classifying male sexual orientation from resting-state EEG.
• Distinct functional EEG patterns were identified in Brodmann areas 40 and 1, differentiating sexual orientation.
• These identified patterns were distinct from those differentiating biological sex.

Conclusions:

• Deep learning can identify electrophysiological trait markers associated with male sexual orientation.
• These findings suggest that sexual orientation has identifiable neurofunctional footprints detectable via EEG.
• The identified EEG patterns are unique to sexual orientation and differ from sex-specific signatures.