Related Experiment Video
Updated: Dec 21, 2025
03:08
Using Human Differentially Expressed Gene Lists to Perform Downstream Pathway Enrichment Analysis and Target Prioritization
Published on: October 3, 2025
809
Exploring the Use of Compound-Induced Transcriptomic Data Generated From Cell Lines to Predict Compound Activity
Benoît Baillif1, Joerg Wichard2, Oscar Méndez-Lucio1,3
1Bayer SAS, Bayer CropScience, Sophia Antipolis, France.
Frontiers in Chemistry
|May 12, 2020
Summary
Predicting drug interactions with molecular targets is challenging. This study shows that gene expression data (transcriptomics) can effectively predict compound activity on targets, outperforming traditional QSAR methods in many cases.
Area of Science:
- Computational chemistry and cheminformatics
- Pharmacology and drug discovery
- Systems biology and transcriptomics
Background:
- Pharmaceuticals and phytopharmaceuticals require specific molecular target interactions for efficacy.
- Identifying unintended molecular interactions (off-targets) is crucial but difficult and costly.
- Quantitative structure-activity relationship (QSAR) models aid in predicting target activity but are limited by data availability for specific targets.
Purpose of the Study:
- To evaluate the utility of large public transcriptomic datasets for predicting compound activity on 69 molecular targets.
- To compare transcriptomic-derived descriptors against traditional QSAR descriptors (Morgan fingerprints) for target prediction.
- To assess the potential of transcriptomics to overcome limitations of chemical space in QSAR modeling.
Main Methods:
- Utilized a large public dataset of compound-induced gene expression measurements (transcriptomics).
- Developed Random Forest models using transcriptomic signatures to predict compound activity on 69 molecular targets.
- Compared model performance against Random Forest models built using Morgan fingerprints (a type of QSAR descriptor).
Main Results:
- Transcriptomic signatures effectively captured compound activity, sometimes revealing similarities even for compounds with different chemical structures.
- Random Forest models based on gene expression signatures performed comparably or better than Morgan fingerprint models for 25% of target prediction tasks.
- Optimal performance was often achieved using transcriptomic data from cell lines exhibiting similar expression signatures for active compounds.
Conclusions:
- Compound-induced transcriptomic data offers a valuable approach for predicting molecular target activity.
- Transcriptomics can overcome the chemical space limitations inherent in traditional QSAR modeling.
- This method presents a promising strategy for drug discovery and chemical de-risking.
