keju: powerful and accurate inference in Massively Parallel Reporter Assays.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Massively Parallel Reporter Assays (MPRAs) are powerful tools for studying gene regulation.
• MPRA experimental designs are complex, leading to uncertainties in DNA and RNA count data.
• Existing statistical methods often fail to account for batch and modality-specific uncertainties.

Purpose of the Study:

• To develop a novel statistical model, keju, for analyzing MPRA data.
• To address the limitations of previous methods in handling uncertainty in MPRA experiments.
• To improve the accuracy and reliability of MPRA data analysis for gene regulatory element interrogation.

Main Methods:

• Developed keju, a hierarchical statistical model for MPRA data analysis.
• Keju models batch-specific and modality-specific uncertainty in RNA counts, conditioned on DNA counts.
• The model estimates transcription rates, differential activity, and promoter composition effects.

Main Results:

• Keju demonstrated significantly improved sensitivity (59%) in simulations compared to existing methods (MPRAnalyze: 31%, BCalm: 9%).
• Keju exhibited lower and more robust false positive rates on real data (6.8% vs. MPRAnalyze: 34%, BCalm: 12%).
• The model effectively handles uncertainties inherent in MPRA experimental designs.

Conclusions:

• Keju offers a statistically rigorous approach to MPRA data analysis.
• The model enhances statistical power and controls false discovery rates, leading to more reliable biological insights.
• Keju represents a significant advancement in the analysis of high-throughput gene regulatory element assays.