MCHMDA:Predicting Microbe-Disease Associations Based on Similarities and Low-Rank Matrix Completion

Area of Science:

• Bioinformatics
• Computational Biology
• Microbiology

Background:

• Microbes are increasingly linked to human diseases like obesity and liver cancer.
• Identifying microbe-disease associations is crucial for understanding disease mechanisms.
• Existing microbe-disease association databases enable computational prediction methods.

Purpose of the Study:

• To develop a computational method for predicting microbe-disease associations.
• To integrate microbial and disease similarities with known associations into a heterogeneous network.
• To evaluate the proposed method's prediction performance against state-of-the-art techniques.

Main Methods:

• Proposed a low-rank matrix completion method (MCHMDA).
• Computed microbe similarity using Gaussian Interaction Profile (GIP) kernel similarity and inhabiting organs.
• Computed disease similarity using GIP, symptom, and functional similarities.
• Constructed a heterogeneous microbe-disease association network.
• Employed the fast Singular Value Thresholding (SVT) algorithm for matrix completion.

Main Results:

• MCHMDA achieved high prediction accuracy in 5-fold Cross Validation (5CV) and Leave-One-Out Cross Validation (LOOCV).
• Achieved AUC values of 0.9251 (5CV) and 0.9495 (LOOCV) on the HMDAD dataset, outperforming other methods.
• Demonstrated prediction generality on an expanded dataset (HMDAD-SUP).
• Case studies validated the practical predictive ability of MCHMDA.

Conclusions:

• MCHMDA is an effective computational method for predicting microbe-disease associations.
• The integration of diverse similarity measures and network construction enhances prediction accuracy.
• MCHMDA shows promise for practical applications in understanding microbe-disease relationships.