Improved drug response prediction by drug target data integration via network-based profiling.

Area of Science:

• Computational biology
• Bioinformatics
• Pharmacogenomics

Background:

• Drug response prediction (DRP) is crucial for personalized medicine, aiming to forecast patient reactions to drugs.
• Current DRP models often use cell line transcriptomes and drug structures but struggle to incorporate drug target interaction (DTI) data due to missing post-treatment transcriptome information.

Purpose of the Study:

• To develop a novel framework that effectively integrates DTI information into existing deep learning-based DRP models.
• To address the challenge of unavailable post-drug-treatment transcriptome data by computing pharmacologic modulation effects.

Main Methods:

• Proposed a framework incorporating NetGP, a module using network propagation to calculate gene perturbation scores.
• NetGP generates a ranked gene list representing perturbation effects, which is processed by a multi-layer perceptron into a fixed-dimension vector.
• This vector is integrated into existing DRP models in a model-agnostic manner.

Main Results:

• The proposed framework significantly boosts the performance of existing DRP models in 64 out of 72 comparisons.
• Performance gains were particularly substantial for predicting responses to unseen drugs, with improvements up to 34% in Pearson's correlation coefficient.

Conclusions:

• The framework successfully leverages DTI information to enhance DRP accuracy, even without direct post-treatment transcriptome data.
• This approach offers a versatile and effective method for improving precision medicine applications through better drug response prediction.