Drug response prediction using graph representation learning and Laplacian feature selection.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Personalized medicine requires accurate prediction of patient drug responses.
• Current experimental methods for drug response prediction are costly and time-consuming.
• Existing computational methods for drug response prediction lack sufficient effectiveness.

Purpose of the Study:

• To develop an effective computational method for predicting cell line-drug responses.
• To leverage integrated biological data for improved drug response prediction.

Main Methods:

• Proposed LGRDRP (Learning Graph Representation for Drug Response Prediction) method.
• Constructed a heterogeneous network integrating cell line miRNA expression, drug chemical structures, gene interactions, and known responses.
• Employed learning graph representation and Laplacian feature selection to extract relevant network topology features.
• Trained a Support Vector Machine (SVM) model using selected features for response prediction.

Main Results:

• LGRDRP demonstrated superior performance compared to state-of-the-art methods.
• Achieved high scores in average area under the ROC curve and precision-recall curve.
• Showcased high recall rate for top-k predicted sensitive cell lines.

Conclusions:

• Integrating multiple data types (cell line and drug information) significantly improves drug response prediction.
• The combination of learning graph representation and Laplacian feature selection enhances prediction accuracy.
• The proposed approach is adaptable for related inference tasks, such as miRNA-disease relationship prediction.