Predicting protein targets for drug-like compounds using transcriptomics.

Area of Science:

• Computational chemistry and drug discovery
• Systems biology and pharmacology

Background:

• Current drug discovery is limited by a focus on known scaffolds, restricting exploration of novel chemical space.
• Identifying targets for new compounds is challenging, especially for those dissimilar to existing drugs.

Purpose of the Study:

• To validate a new computational pipeline for identifying small molecule-protein interactions, including for compounds lacking similarity to known drugs.
• To overcome the bias towards known scaffolds in drug discovery and expand the exploration of chemical space.

Main Methods:

• Analyzed differential mRNA profiles from cell types exposed to drugs and after gene knockdowns (KD).
• Applied supervised machine learning to correlate drug-induced gene regulatory networks with KD signatures.
• Enriched predictions using structure-based screening and validated interactions with difficult-to-modulate targets.

Main Results:

• Achieved top-10/top-100 target prediction accuracies of 26%/41% on a validation set of 152 FDA-approved drugs and 3104 potential targets.
• Successfully predicted and validated chemical interactors for 1680 compounds, including those targeting HRAS and KRAS.
• Demonstrated that drug-target interactions manifest as gene expression correlations between drug treatment and target gene KD.

Conclusions:

• The validated pipeline effectively identifies small molecule-protein interactions, even for novel compounds and weak binders.
• This approach offers new insights into cellular responses to disrupted protein interactions and complex drug phenotypes.
• The pipeline has the potential to accelerate the discovery and development of novel chemical classes for therapeutic targets.