Probiotic-Disease Association Prediction via Cross-Modal Feature Aggregation

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Probiotics offer health benefits and can complement traditional medicine.
• Current methods for identifying probiotic-disease links are inefficient and labor-intensive.
• Existing computational approaches often neglect crucial probiotic and disease characteristics and dataset noise.

Purpose of the Study:

• To develop an efficient computational framework for predicting probiotic-disease associations.
• To address limitations of existing methods by incorporating multi-source features and deep learning.
• To present MFFPDA, the first deep-learning framework for multi-feature fusion in probiotic-disease association prediction.

Main Methods:

• Systematic screening of a probiotic-disease association dataset.
• Collection of diverse probiotic and disease-related data.
• Calculation and integration of multiple probiotic and disease features using feature extraction and fusion modules within a deep-learning framework (MFFPDA).

Main Results:

• MFFPDA demonstrated superior performance compared to all other evaluated methods.
• Feature visualization confirmed the importance and validity of using multi-source features.
• Case studies on colonic pseudo-obstruction and dysentery validated MFFPDA's predictive accuracy and reliability.

Conclusions:

• MFFPDA offers a reliable and effective computational tool for predicting probiotic-disease associations.
• The integration of multi-source features significantly enhances prediction accuracy.
• This framework provides a valuable alternative to traditional experimental screening methods.