An Attention-Based Deep Learning Method for Schizophrenia Patients Classification Using DNA Methylation Data.

Area of Science:

• Genetics
• Bioinformatics
• Computational Neuroscience

Background:

• Schizophrenia classification is challenging due to high-dimensional, non-Gaussian DNA methylation data with small sample sizes.
• Exploring the relationship between DNA characteristics and schizophrenia is crucial for understanding the disease.
• Existing feature selection methods struggle with the complex nature of this data.

Purpose of the Study:

• To develop an effective deep learning-based classification method for schizophrenia using DNA methylation data.
• To address the challenges posed by high dimensionality and non-Gaussian distributions in DNA methylation datasets.
• To improve the accuracy of classifying schizophrenia patients from healthy controls.

Main Methods:

• A novel feature selection method employing an attention mechanism with a weight gated layer was designed.
• This method generates a task-related sparse representation of DNA methylation data.
• The approach integrates feature selection directly within a deep learning network structure.

Main Results:

• The proposed method demonstrated superior performance compared to existing feature selection techniques.
• High classification accuracy was achieved on a real-world dataset for distinguishing schizophrenia patients.
• The attention mechanism effectively identified relevant DNA methylation features for classification.

Conclusions:

• Deep learning with attention-based feature selection offers a promising approach for schizophrenia classification using DNA methylation data.
• The developed method successfully overcomes limitations of traditional methods in handling complex biological data.
• This work contributes to a better understanding of DNA methylation's role in schizophrenia.