X-LDA: An interpretable and knowledge-informed heterogeneous graph learning framework for LncRNA-disease association...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Long noncoding RNAs (lncRNAs) play crucial roles in gene expression regulation and epigenetic modifications.
• Identifying lncRNA-disease associations (LDAs) is vital for understanding disease mechanisms.
• Existing computational methods for LDAs often lack interpretability, hindering biological and medical researcher adoption.

Purpose of the Study:

• To propose X-LDA, an interpretable and knowledge-informed heterogeneous graph learning framework for predicting LDAs.
• To provide intuitive explanations for LDA predictions, enhancing trust and understanding.
• To improve upon the performance and interpretability of current LDA prediction methods.

Main Methods:

• Constructed a knowledge-informed heterogeneous graph integrating LDAs, lncRNA similarities, and disease similarities.
• Defined nine graph patch types to capture topological relationships for interpretable predictions.
• Employed graph patch convolution with parameter sharing and multi-convolution kernels for feature extraction and context embedding.
• Utilized integrated gradients for post-hoc explanations of LDA predictions.

Main Results:

• X-LDA demonstrated superior performance compared to nine state-of-the-art methods.
• Achieved an average area under the receiver operating curve (AUC) of 0.9891 and an average area under the precision-recall curve (AUPRC) of 0.7907.
• Ablation studies and interpretability experiments confirmed X-LDA's robustness, learnability, predictability, and interpretability.
• Case studies on prostate, colorectal, and breast cancers showcased X-LDA's practical applicability.

Conclusions:

• X-LDA offers a robust and interpretable approach for predicting lncRNA-disease associations.
• The framework enhances biological insight into gene regulation and disease.
• X-LDA's performance and interpretability make it a valuable tool for researchers in genomics and medicine.