Transcriptomic data and biomedical literature synergize in finding pharmacologic gene regulators.

Area of Science:

• Genomics
• Computational Biology
• Pharmacology

Background:

• Mendelian disorders often lack targeted therapies due to single-gene defects.
• Gene expression perturbations (overexpression, knockout, knockdown) can model these disorders.
• RNA-Sequencing (RNA-Seq) data holds potential for identifying counteracting drugs, but manual annotation is a bottleneck.

Purpose of the Study:

• To introduce SNACKKSS (Signature-based Networks from Automatically Curated Knockout, Knockdown, and Small-molecule Studies).
• To automate the curation of gene disruption and drug studies from public datasets.
• To improve drug prioritization for Mendelian disorders by integrating RNA-Seq data.

Main Methods:

• Automated curation of gene expression datasets from the Gene Expression Omnibus.
• Integration of RNA-Seq data with uniformly computed read counts.
• Development of regulatory relationship prediction models, including SNACKKSS and its variant SA4.
• Cross-validation to assess prediction accuracy and contribution to drug discovery.

Main Results:

• SNACKKSS successfully automates the curation and analysis of large-scale gene expression data.
• The SA4 variant of SNACKKSS uniquely contributes to identifying protein-inhibiting compounds.
• Ensembling SNACKKSS with other predictive tools enhances drug-repurposing capabilities.
• Machine learning model performance can vary across different computational devices, necessitating careful testing.

Conclusions:

• SNACKKSS offers a powerful, automated approach to leverage RNA-Seq data for drug discovery in genetic disorders.
• Integrating diverse data sources, including RNA-Seq, significantly improves drug-repurposing screens.
• Researchers should be mindful of computational reproducibility when using complex machine learning models.