DMFMDA: Prediction of Microbe-Disease Associations Based on Deep Matrix Factorization Using Bayesian Personalized...

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Understanding microbe-disease associations is crucial for elucidating disease pathogenesis.
• Existing methods may not fully leverage complex biological data for accurate predictions.

Purpose of the Study:

• To develop and validate a novel deep learning model for predicting microbe-disease associations.
• To improve the accuracy and reliability of microbe-disease association predictions.

Main Methods:

• A deep matrix factorization prediction model (DMFMDA) was developed using deep neural networks.
• The model incorporates embedding layers for diseases and microbes, matrix factorization, and multi-layer perceptrons.
• Bayesian Personalized Ranking was employed for model optimization, utilizing unobserved data.

Main Results:

• DMFMDA achieved high average AUCs of 0.9091 (5-fold cross-validation) and 0.9103 (Leave-one-out cross-validation).
• The model outperformed three state-of-the-art methods in prediction accuracy.
• Case studies showed high validation rates for predicted microbe-disease associations for asthma, inflammatory bowel disease, and colon cancer.

Conclusions:

• DMFMDA demonstrates a successful application of deep learning in predicting microbe-disease associations.
• The model offers a promising approach for advancing our understanding of microbial roles in human diseases.