CheckSumStats

CheckSumStats identifies and corrects metadata errors and analytical issues in genome-wide association study (GWAS) summary statistics to improve data integrity for downstream analyses including Mendelian randomization.

Key Features:

• Implementation: Provided as an R package for programmatic quality control of GWAS summary statistics.
• Error Identification: Systematically identifies metadata errors by comparing study-specific statistics against external reference datasets including the NHGRI-EBI GWAS Catalog and 1000 Genomes super-populations.
• Analytical Issue Detection: Detects analytical issues by evaluating reported genetic effect sizes against expected values using three sets of variants: GWAS hits for fatty acids, GWAS hits for cancer, and a reference set from the 1000 Genomes project.
• Comprehensive Quality Control Pipeline: Implements a QC pipeline developed within the Fatty Acids in Cancer Mendelian Randomization Collaboration (FAMRC) that resolves analytical issues and excludes unreliable data.
• Summary Data Collation: Collates summary-level GWAS data from multiple studies to enable cross-study comparisons and QC.

Scientific Applications:

• Mendelian randomization quality control: Identifies metadata inaccuracies such as incorrect effect allele columns that could bias Mendelian randomization analyses.
• Large-scale GWAS summary data QC: Applied to collate and quality-control summary data from six fatty acid studies and 49 cancer GWAS, identifying metadata errors and analytical issues in 13% of studies.
• Consortium-level data harmonization: Used within the FAMRC to improve integrity of collated summary data prior to downstream post-GWAS analyses.

Methodology:

Summary-level GWAS data are collated; metadata errors are identified by comparison with external references (NHGRI-EBI GWAS Catalog, 1000 Genomes super-populations); analytical issues are detected by comparing reported genetic effect sizes to expected values using specified variant sets (GWAS hits for fatty acids, GWAS hits for cancer, and a 1000 Genomes reference set); identified issues are resolved and unreliable data excluded.