FC1000: normalized gene expression changes of systematically perturbed human cells.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Systematic study of transcriptional responses in human cells is emerging.
• The L1000 compendium is the largest dataset of gene expression profiles from treated human cells.
• Large datasets present significant data normalization challenges.

Purpose of the Study:

• To develop a novel and practical approach for accurate fold change estimation from the L1000 compendium.
• To address the data normalization challenges inherent in large-scale gene expression datasets.
• To enhance the utility of the L1000 data for biomedical data mining.

Main Methods:

• Utilized the RUV (Remove Unwanted Variation) statistical framework, extended for big data.
• Developed an estimation procedure tuning the RUV model using dataset-specific statistical measures.
• Employed evaluation endpoints, including p-value distributions and gene knockdown controls, for feedback.
• Applied metrics to disjoint data splits and integrated results for optimal normalization.

Main Results:

• Successfully reduced bias and noise in the L1000 gene expression data.
• Developed a practical pipeline for normalizing large-scale transcriptional response data.
• The normalization procedure enhances the potential for pharmacological and functional genomic analyses.

Conclusions:

• The novel normalization approach significantly improves the quality of the L1000 compendium.
• This method makes the L1000 dataset more amenable to robust data mining.
• The developed R package facilitates broader application of the L1000 resource in biomedical research.