miRTMC

miRTMC predicts microRNA (miRNA) targets by applying a matrix completion algorithm to integrate miRNA similarity, gene similarity, and miRNA-gene interaction networks and thereby improve prediction accuracy and reduce false positives.

Key Features:

• Matrix Completion Algorithm: Employs a matrix completion approach that assumes the adjacency matrix of an integrated heterogeneous network is low-rank to recover missing miRNA-gene associations.
• Heterogeneous Network Integration: Integrates the miRNA similarity network, the gene similarity network, and the miRNA-gene interaction network into a single heterogeneous network for holistic prediction.
• Nuclear Norm Regularization: Solves the matrix completion problem via a nuclear norm regularized linear least squares model under non-negative constraints to manage sparsity and noise.
• Alternating Direction Method of Multipliers (ADMM): Uses ADMM to numerically optimize the regularized matrix completion problem for efficient convergence.

Scientific Applications:

• Studying disease mechanisms: Predicts miRNA targets to investigate miRNA dysfunctions that contribute to human diseases.
• Therapeutic target identification: Reveals miRNA-gene regulatory relationships that can inform identification of potential therapeutic targets.
• Method benchmarking: Provides predictions that can outperform existing methods in various evaluation metrics, enabling comparative assessment of miRNA target prediction approaches.

Methodology:

Integrates known experimentally validated miRNA targets into a heterogeneous network and applies matrix completion with a low-rank adjacency assumption using a nuclear norm regularized linear least squares model under non-negative constraints, optimized numerically by ADMM.