Turning Vice into Virtue: Using Batch-Effects to Detect Errors in Large Genomic Data Sets.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Large-scale genomic studies often require combining data from diverse sequencing centers and platforms.
• Heterogeneity in data generation introduces systematic errors, complicating variant detection and analysis.
• Existing methods struggle to comprehensively identify and correct these batch effects.

Purpose of the Study:

• To develop and validate a novel computational method for detecting systematic errors in combined genomic datasets.
• To quantify the prevalence of such errors, particularly in coding versus non-coding regions.
• To demonstrate the utility of batch effects for identifying data quality issues.

Main Methods:

• Devised a method analyzing pairs of variants on different chromosomes that co-occur within individuals.
• Studied the abundance of these variant pairs across different genomes to identify systematic errors (batch effects).
• Applied the method to the 1000 Genomes dataset and compared findings with data from different sequencing technologies.

Main Results:

• Identified systematic errors enriched in coding regions of the 1000 Genomes dataset, affecting ~1% of high-frequency variants.
• Errors outside coding regions were significantly rarer (<0.001%).
• Predicted errors were less frequent in data from different sequencing technologies, supporting their validity, and were observed in other large datasets.

Conclusions:

• The developed method effectively detects systematic errors arising from combining diverse genomic data.
• Batch effects, often viewed as a nuisance, can be leveraged to improve genomic data quality.
• Findings highlight the importance of accounting for data generation variability in large-scale genomic analyses.