DeepDRA: Drug repurposing using multi-omics data integration with autoencoders.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in oncology

Background:

• Cancer treatment faces challenges with costly and time-consuming drug development.
• Drug repurposing offers a promising avenue, with machine learning showing potential.
• Deep learning methods surpass classical machine learning in predicting cancer drug responses.

Purpose of the Study:

• To develop a deep learning model for predicting cancer drug response.
• To utilize multi-omics data, drug descriptors, and fingerprints for enhanced prediction.
• To facilitate drug repurposing for novel cancer therapeutics.

Main Methods:

• Developed a deep learning model integrating autoencoders for dimensionality reduction of multi-omics data.
• Employed a multi-task learning approach connecting autoencoders with Multi-Layer Perceptrons (MLPs).
• Validated the model on three large datasets: Genomics of Drug Sensitivity in Cancer (GDSC), Cancer Therapeutics Response Portal (CTRP), and Cancer Cell Line Encyclopedia (CCLE).

Main Results:

• The model consistently outperformed state-of-the-art methods across datasets, achieving an Area Under the Precision-Recall Curve (AUPRC) of 0.99.
• In cross-dataset evaluations (e.g., training on GDSC, testing on CCLE), the model achieved an AUPRC of 0.72, surpassing previous works.
• Demonstrated superior generalization capabilities compared to existing models.

Conclusions:

• Presented a novel deep learning model for accurate cancer drug response prediction.
• The model facilitates efficient drug repurposing, potentially accelerating the discovery of new cancer drugs.
• Highlights the significant potential of advanced deep learning in advancing precision cancer therapeutics.