Predicting alcohol use disorder remission: a longitudinal multimodal multi-featured machine learning approach.

Area of Science:

• Neuroscience
• Genetics
• Psychiatry

Background:

• Alcohol use disorder (AUD) poses significant treatment challenges.
• Identifying biomarkers for AUD recovery can improve treatment outcomes and reduce costs.
• Previous research often lacks diverse ancestry and multimodal data integration.

Purpose of the Study:

• To develop and validate predictive models for alcohol use disorder (AUD) recovery.
• To identify predisposition biomarkers associated with AUD remission.
• To explore the impact of sex and ancestry on prediction accuracy.

Main Methods:

• Utilized multimodal machine learning on a dataset of 1376 individuals (European and African ancestry) from the Collaborative Study on the Genetics of Alcoholism (COGA).
• Included features such as electroencephalography (EEG) source-level functional brain connectivity, Polygenic Risk Scores (PRS), medications, and demographics.
• Performed sex and ancestry-stratified analyses using supervised linear Support Vector Machine, differentiating ancestry by self-report and genetic data.

Main Results:

• Multifeatured models outperformed single-domain models in predicting AUD recovery.
• The highest accuracy (86.04%) was achieved in African American males using PRS, EEG connectivity, marital, and employment status.
• Identified key predictive features including PRS for neuroticism, depression, and aggression, education, employment, medication use, and specific brain connectivity patterns (e.g., default mode network, insula).

Conclusions:

• Machine learning models integrating diverse data types show high accuracy in predicting AUD remission.
• Sex and ancestry-specific features are crucial for enhancing prediction models.
• Identified biomarkers offer potential for targeted addiction treatment strategies and personalized medicine approaches for AUD recovery.