Development of Drug-Induced Gene Expression Ranking Analysis (DIGERA) and Its Application to Virtual Screening for...

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Understanding drug-target interactions is key for drug discovery and precision medicine.
• Phenotype-based drug discovery using gene expression is effective but limited by experimental costs.
• Predicting drug-induced gene expression profiles computationally is needed for large-scale screening.

Purpose of the Study:

• To develop a computational framework, DIGERA, for predicting drug-induced gene expression rankings.
• To create novel numerical features to improve prediction accuracy across cell lines.
• To apply DIGERA for de novo design of potential drug compounds, specifically PARP1 inhibitors.

Main Methods:

• Developed DIGERA, a Lasso-based ensemble framework using LINCS L1000 data.
• Created novel numerical features for chemicals, cell lines, and experimental conditions.
• Integrated DIGERA with iterative fine-tuning for de novo drug design.

Main Results:

• DIGERA demonstrated superior performance in predicting gene expression rankings compared to baseline models (F1@K metric).
• The framework successfully predicted 10 PARP1 inhibitors with favorable drug-like properties.
• Nine of the suggested compounds were novel, with six analogs linked to PARP1 inhibition.

Conclusions:

• DIGERA enhances model performance and robustness through novel features and ensemble learning.
• The framework effectively aids in virtual screening for novel PARP1 inhibitors.
• DIGERA shows significant potential for accelerating drug discovery and development.