Deep-Learning-Based Integration of Sequence and Structure Information for Efficiently Predicting miRNA-Drug...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are critical regulators in cancer progression and drug resistance.
• Targeting miRNAs offers a promising strategy for overcoming cancer treatment resistance.
• Computational prediction of miRNA-drug associations (MDAs) is underexplored, with existing methods limited by interaction graph density.

Purpose of the Study:

• To develop a novel deep learning method for predicting miRNA-drug associations (MDAs).
• To integrate sequence and structural information for more accurate MDA prediction.
• To address limitations of existing graph-based computational approaches.

Main Methods:

• Proposed DLST-MDA, a deep learning model integrating miRNA sequence and drug sequence/structure information.
• Utilized multiscale convolutional neural networks (CNNs) for learning sequence embeddings.
• Employed graph neural networks (GNNs) to capture structural features of drug molecules.
• Constructed a benchmark dataset for rigorous evaluation.

Main Results:

• DLST-MDA demonstrated superior performance compared to state-of-the-art methods in predicting MDAs.
• The model effectively leverages attribute information, moving beyond reliance on interaction graphs.
• Case studies confirmed the utility of DLST-MDA in identifying potential novel miRNA-drug associations for anticancer drugs.

Conclusions:

• DLST-MDA represents a significant advancement in computational prediction of miRNA-drug associations.
• The integration of sequence and structural data enhances prediction accuracy.
• This method holds promise for discovering new therapeutic strategies against cancer drug resistance.