Assessing reproducibility of high-throughput experiments in the case of missing data

Area of Science:

• Biological and Biomedical Research
• Genomics
• Computational Biology

Background:

• High-throughput experiments generate substantial missing data, particularly in single-cell RNA sequencing (scRNA-seq), due to low detection levels.
• Excluding missing data from reproducibility assessments can lead to inaccurate conclusions about experimental reliability.

Purpose of the Study:

• To develop a robust regression model for assessing experimental reproducibility in the presence of extensive missing data.
• To investigate the impact of operational factors, such as sequencing depth and platform, on reproducibility when data is incomplete.

Main Methods:

• Developed a latent variable regression model extending correspondence curve regression to handle missing values.
• Incorporated a latent variable approach to model missing observations within reproducibility assessments.
• Validated the method using simulations and a real-world single-cell RNA-seq dataset.

Main Results:

• The proposed method demonstrated higher accuracy in detecting reproducibility differences compared to existing measures.
• The model effectively assessed how operational factors influence reproducibility with missing data.
• Identified cost-effective sequencing depths for achieving adequate reproducibility in scRNA-seq.

Conclusions:

• The novel regression model provides a more reliable assessment of high-throughput experiment reproducibility, especially when dealing with missing data.
• This approach enhances the understanding of how experimental design choices impact data reliability.
• Offers practical insights for optimizing experimental parameters in scRNA-seq and similar high-throughput studies.