Empowering Graph Neural Networks with Block-Based Dual Adaptive Deep Adjustment for Drug Resistance-Related NcRNA...

Area of Science:

• Computational biology
• Genomics
• Cancer research

Background:

• Drug resistance to chemotherapy is a major obstacle in cancer treatment.
• Noncoding RNAs (ncRNAs) play a significant role in mediating cancer drug resistance.
• Traditional experimental methods for identifying ncRNA-drug resistance associations are time-consuming and labor-intensive.

Purpose of the Study:

• To develop an efficient computational framework, B-NDRA, for predicting ncRNA-drug resistance associations (NDRA).
• To overcome the limitations of existing models by incorporating similarity information between ncRNAs and drug resistance.
• To facilitate the discovery of novel ncRNA targets for overcoming cancer drug resistance.

Main Methods:

• Constructed a heterogeneous graph integrating known ncRNA-drug resistance pairs and similarity fusion information.
• Employed an attention mechanism for local feature aggregation and dimensionality reduction.
• Utilized a graph neural network (GNN) for learning global node embeddings.
• Integrated dual adaptive deep adjustment architectures for feature extraction and balancing.
• Applied a multilayer perceptron for final NDRA prediction.

Main Results:

• B-NDRA achieved high performance in 5-fold cross-validation with average AUC of 92.2% and AUPR of 91.9%.
• Comparative evaluations showed B-NDRA outperformed existing models like GAEMDA, GRPAMDA, and LRGCPND across multiple metrics.
• Case studies on Doxorubicin and Imatinib demonstrated B-NDRA's practical utility in identifying potential NDRA.

Conclusions:

• B-NDRA is a powerful computational tool for discovering ncRNA-drug resistance associations.
• The framework's ability to integrate similarity information enhances prediction accuracy.
• B-NDRA holds significant potential for advancing cancer research and therapeutic development by aiding in overcoming drug resistance.