Drug Response Prediction as a Link Prediction Problem.

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Drug response prediction is crucial for precision medicine, particularly in cancer research.
• Genomic data from cell lines are commonly used for machine learning models, but integrating molecular network data presents challenges due to increased dimensionality.
• Existing methods struggle to effectively incorporate network information for robust drug response prediction.

Purpose of the Study:

• To develop a novel machine learning approach for accurate and reproducible drug response prediction.
• To overcome the challenges of high dimensionality when integrating molecular network data into predictive models.
• To leverage network information for enhanced prediction of drug sensitivity and resistance in cancer cell lines.

Main Methods:

• Formulated drug response prediction as a link prediction problem using a heterogeneous network.
• Represented drug response data for a large cohort of cell lines within this network.
• Computed "network profiles" for cell lines and drugs to predict links between them.

Main Results:

• Achieved 85% accuracy in classifying sensitive and resistant cell line-drug pairs.
• Demonstrated accurate and reproducible classification through leave-one-out cross-validation and cross-classification on independent datasets.
• Examined the biological relevance of the computed network profiles, indicating their interpretability.

Conclusions:

• The proposed network-based link prediction approach effectively predicts drug response in cancer cell lines.
• This method offers a robust and reproducible solution for precision medicine by integrating molecular network data.
• The network profiles provide biologically relevant insights into drug-target interactions and cellular responses.