MNNMDA

MNNMDA predicts potential associations between microbes and diseases by minimizing the matrix nuclear norm to perform low-rank matrix completion for microbe-disease association inference.

Key Features:

• Matrix Nuclear Norm Minimization: Formulates microbe-disease association prediction as a low-rank matrix completion task using matrix nuclear norm minimization with additional regularization terms.
• Gaussian Interaction Profile Kernel Similarity: Computes Gaussian interaction profile kernel similarity for both diseases and microbes.
• Functional Similarity: Incorporates functional similarity measures between microbial and disease entities.
• Heterogeneous Information Network Construction: Integrates an integrated disease similarity network, an integrated microbe similarity network, and a known microbe-disease bipartite network into a heterogeneous information network.
• Performance Evaluation: Evaluated on HMDAD, Disbiome, and Combined Data using AUROC and AUPR under 5-fold cross-validation, achieving AUROCs of 0.9536 (HMDAD) and 0.9364 (Disbiome) and outperforming KATZHMDA, LRLSHMDA, NTSHMDA, GATMDA, and KGNMDA.
• Case Studies: Validated through case studies on colon cancer and inflammatory bowel disease (IBD).

Scientific Applications:

• Pathological mechanism discovery: Predicts microbe-disease associations to support investigation of disease-related microbial mechanisms.
• Therapeutic target prioritization: Aids in identifying and prioritizing disease-associated microbes as potential targets for intervention.
• Microbial interaction analysis: Supports analysis of complex microbial interactions within the human body.

Methodology:

MNNMDA applies matrix nuclear norm minimization for low-rank matrix completion with regularization terms, computes Gaussian interaction profile kernel similarity for diseases and microbes and functional similarity, constructs a heterogeneous information network by integrating disease similarity, microbe similarity, and a known microbe-disease bipartite network, and evaluates performance via 5-fold cross-validation using AUROC and AUPR on HMDAD, Disbiome, and Combined Data.