DeepCOP: deep learning-based approach to predict gene regulating effects of small molecules.

Area of Science:

• Bioinformatics and chemogenomics
• Computational biology
• Drug discovery

Background:

• Bioinformatics and chemogenomics advances are accelerating the discovery of small molecule regulators.
• Deep learning combined with large genomic and molecular data can revolutionize predictive biology.

Purpose of the Study:

• To present Deep gene COmpound Profiler (DeepCOP), a deep learning model predicting gene-regulating effects of compounds.
• To enable direct identification of drug candidates causing desired gene expression responses, independent of protein target interactions.

Main Methods:

• Combined molecular fingerprint descriptors and gene ontology-derived gene descriptors.
• Trained deep neural networks on differential gene regulation data from the LINCS database.
• Validated models using an external RNA-Seq dataset on antiandrogen effects in prostate cancer cells.

Main Results:

• Achieved 10-fold cross-validation ROC AUC scores above 0.80 and enrichment factors >5.
• Demonstrated effective synergy between molecular and genomic descriptors using deep learning.
• Successfully screened novel drug candidates with desired gene expression effects.

Conclusions:

• Deep learning models can effectively predict compound-induced gene expression changes.
• DeepCOP facilitates the screening of novel drug candidates for desired gene expression effects.
• This approach holds promise for developing cancer therapeutics and advancing precision oncology.