Discrimination of alcohol dependence based on the convolutional neural network

Area of Science:

• Genetics and Neuroscience
• Machine Learning in Healthcare
• Addiction Research

Background:

• Alcohol dependence (AD) poses significant public health challenges.
• Identifying reliable biomarkers for AD is crucial for timely intervention.
• Genetic factors, specifically single nucleotide polymorphisms (SNPs), are implicated in AD susceptibility.

Purpose of the Study:

• To investigate the efficacy of combining genetic data (SNPs in HTR3A and HTR3B genes) with demographic information for discriminating between individuals with and without alcohol dependence.
• To compare the performance of different machine learning models, including Grid Search-Support Vector Machine (GS-SVM), Convolutional Neural Network (CNN), and CNN combined with Long Short-Term Memory (CNN-LSTM), in classifying alcohol dependence.

Main Methods:

• Genomic DNA was analyzed for 20 single nucleotide polymorphisms (SNPs) in the HTR3A and HTR3B genes.
• Feature fusion was performed by integrating SNP data with demographic variables (age, education, marital status).
• Machine learning algorithms (GS-SVM, CNN, CNN-LSTM) were employed for classification tasks.

Main Results:

• The combination of 19 SNPs and academic qualifications demonstrated the strongest discriminative power.
• The Grid Search-Support Vector Machine (GS-SVM) achieved an Area Under the ROC Curve (AUC) of 0.87.
• The CNN-LSTM model yielded an AUC of 0.88, while the CNN model achieved the highest AUC of 0.92.

Conclusions:

• The Convolutional Neural Network (CNN) model exhibits superior performance in discriminating alcohol dependence compared to traditional Support Vector Machine (SVM) approaches.
• Integrating genetic markers with demographic data enhances the accuracy of AD classification.
• These findings suggest that CNN models hold significant potential for the early and accurate diagnosis of alcohol dependence, facilitating timely treatment.