Drug Repurposing Using Deep Embeddings of Gene Expression Profiles.

Area of Science:

• Computational biology
• Pharmacogenomics
• Machine learning in drug discovery

Background:

• Accurate assessment of compound functional similarity is crucial for computational drug repositioning.
• Existing methods often struggle with noisy gene expression data and require extensive pharmacological information.

Purpose of the Study:

• To develop a novel method for measuring compound functional similarity using deep neural networks and gene expression data.
• To improve the accuracy and efficiency of drug repositioning by leveraging unlabeled gene expression data.

Main Methods:

• Utilized deep neural networks to learn a denoised embedding from gene expression data.
• Employed an unsupervised learning approach, requiring only compound identity labels.
• Compared the performance of the learned embedding space similarity with previous similarity measures.

Main Results:

• The deep learning embedding substantially reduced noise in gene expression data, increasing similarity among compound replicates.
• Similarity in the learned embedding space accurately predicted pharmacological similarities without explicit pharmacological labels.
• Achieved superior performance compared to existing similarity measures for gene expression data.
• Successfully identified structurally dissimilar compounds with shared therapeutic and biological targets.

Conclusions:

• The developed method provides an improved engine for drug repurposing based on gene expression data.
• This approach can uncover novel functional relationships between compounds, facilitating drug discovery.
• The tool, DeepCodex, is available online to support drug repositioning efforts.