Identifying lncRNA-disease association based on GAT multiple-operator aggregation and inductive matrix completion.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Computable models are crucial for inferring long non-coding RNA-disease associations (LDA), but data sparsity hinders accuracy.
• Existing heterogeneous biological data presents challenges for traditional computational approaches in LDA prediction.

Purpose of the Study:

• To propose a novel computational model, MM-LDA, for enhanced LDA prediction.
• To address data sparsity and improve the accuracy and efficiency of LDA inference.

Main Methods:

• Developed MM-LDA by integrating Graph Attention Network (GAT) with multiple-operator aggregation and Inductive Matrix Completion (IMC).
• GAT enhances lncRNA and disease node features; IMC reconstructs the LDA network to handle data deficiencies and cold-start problems.
• Utilized the Adam optimizer for adaptive learning rate adjustment to improve convergence speed and avoid local optima.

Main Results:

• MM-LDA demonstrated superior predictive ability, achieving an AUC of 0.9395 and AUPR of 0.8057 via 5-fold cross-validation.
• The model exhibited a 6.45% lower time cost compared to the advanced GAMCLDA model.
• Achieved comprehensive prediction performance, balancing accuracy and computational efficiency.

Conclusions:

• MM-LDA effectively overcomes data sparsity challenges in LDA prediction.
• The proposed model offers a significant advancement in predictive accuracy and computational efficiency for lncRNA-disease association studies.
• MM-LDA provides a robust framework for future research in computational biology and disease association.