Transfer Learning With BioBERT Embeddings for lncRNA-Disease Association Prediction

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Long non-coding RNAs (lncRNAs) regulate gene expression and are implicated in diseases like cancer and neurological disorders.
• Understanding lncRNA-disease links is crucial for developing biomarkers and therapies.
• Traditional experimental methods for lncRNA-disease association analysis are limited in scalability and cost.

Purpose of the Study:

• To develop a novel computational framework, TBLDA, for predicting long non-coding RNA-disease associations.
• To leverage transfer learning and advanced natural language processing techniques for improved prediction accuracy.
• To provide a powerful tool for identifying disease-associated lncRNAs to aid medical research.

Main Methods:

• Developed TBLDA, a transfer learning-based framework utilizing BERT-derived embeddings for disease term semantics.
• Integrated a pre-trained miRNA-disease association model to transfer knowledge.
• Employed leave-one-out cross-validation (LOOCV) and 5-fold cross-validation (5-fold CV) for performance evaluation.

Main Results:

• TBLDA achieved high predictive performance with AUC scores of 0.9801 (LOOCV) and 0.9721 (5-fold CV).
• The framework significantly outperformed five competitive baseline models.
• Demonstrated the effectiveness of the transfer learning strategy in enhancing lncRNA-disease prediction.

Conclusions:

• TBLDA is a robust and effective computational tool for predicting lncRNA-disease associations.
• The transfer learning approach enhances predictive accuracy, surpassing existing methods.
• TBLDA can complement experimental research by identifying potential disease-associated lncRNAs, facilitating further medical studies.