Exploring noncoding RNAs in thyroid cancer using a graph convolutional network approach.

Area of Science:

• Genomics
• Bioinformatics
• Oncology

Background:

• Noncoding RNAs (ncRNAs) play a critical role in the initiation and progression of thyroid cancer.
• Current wet-lab experimental methods for studying ncRNA-thyroid cancer associations are costly and difficult to scale.
• Existing databases lack comprehensive data specific to ncRNA-thyroid cancer interactions, highlighting a need for computational solutions.

Purpose of the Study:

• To develop and evaluate a computational approach for predicting associations between ncRNAs and thyroid cancer.
• To address the limitations of experimental methods and data availability in ncRNA-thyroid cancer research.

Main Methods:

• A linear residual graph convolution network, TCGCN, was developed to predict ncRNA-thyroid cancer associations.
• A bipartite graph was constructed using extensive ncRNA-disease association data.
• The model employs linear embedding propagation and a weighted sum of embeddings across convolutional layers for prediction.

Main Results:

• TCGCN achieved high performance in 5-fold cross-validation on the ncRNA-thyroid cancer dataset.
• The model demonstrated superior performance compared to state-of-the-art methods, with an AUC of 0.8162 and an AUPR of 0.8049.
• Case studies confirmed the practical utility and effectiveness of the TCGCN method.

Conclusions:

• TCGCN provides a powerful and efficient computational tool for predicting ncRNA-thyroid cancer associations.
• This method can aid in the diagnosis and treatment strategies for thyroid cancer by elucidating ncRNA involvement.
• The study underscores the potential of graph-based deep learning models in advancing cancer research.