A self-knowledge distillation-driven CNN-LSTM model for predicting disease outcomes using longitudinal microbiome...

Area of Science:

• Microbiome research
• Computational biology
• Artificial intelligence in medicine

Background:

• The human microbiome is complex and dynamic, offering valuable temporal insights for disease prediction.
• Analyzing longitudinal microbiome data is challenging due to missing data and heterogeneity.
• Current methods struggle to fully capture dynamic microbiome patterns for accurate disease outcome prediction.

Purpose of the Study:

• To develop an efficient hybrid deep learning model for analyzing longitudinal microbiome profiles.
• To accurately predict disease outcomes using dynamic microbiome data.
• To improve upon existing temporal deep learning models for microbiome analysis.

Main Methods:

• Proposed a hybrid deep learning architecture combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM).
• Integrated self-knowledge distillation to enhance model accuracy.
• Applied the model to longitudinal microbiome datasets from the PROTECT and DIABIMMUNE studies.

Main Results:

• Achieved significant improvements in Area Under the Receiver Operating Characteristic Curve (AUC) scores.
• Obtained AUC scores of 0.889 for the PROTECT study and 0.798 for the DIABIMMUNE study.
• Demonstrated superior performance compared to state-of-the-art temporal deep learning models.

Conclusions:

• Developed an effective AI-based tool for predicting disease outcomes from longitudinal microbiome data.
• The proposed model offers a powerful approach for leveraging dynamic microbiome information.
• Findings highlight the potential of advanced AI in personalized medicine and disease risk assessment.