Fast score test with global null estimation regardless of missing genotypes.

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Genome-wide association studies (GWASs) commonly use logistic regression for binary traits.
• Existing methods like Wald, likelihood ratio, and conventional score tests (CST) face computational challenges with missing genotype data.
• Maximum likelihood estimation (MLE) in Wald and likelihood ratio tests is computationally intensive for each SNP.

Purpose of the Study:

• To develop novel score tests (PM1 and PM2) that enhance computational efficiency in GWASs, particularly with missing genotype data.
• To provide a faster alternative to existing association tests by utilizing a single global null estimator.
• To evaluate the performance and applicability of the proposed score tests.

Main Methods:

• Proposed two new score tests, PM1 and PM2, employing a single global null estimator for all SNPs, irrespective of missing genotypes.
• Conducted theoretical analysis to compare the asymptotic power of PM1 and PM2 with CST.
• Performed simulation studies and applied the methods to real GWAS data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Main Results:

• PM2 demonstrates equivalent asymptotic power to CST, while PM1 has asymptotically lower power.
• The proposed score tests (PM1 and PM2) significantly improve computational speed, ranging from 6 to 18 times faster than existing tests on ADNI-GWAS data.
• Type I error rates and power were evaluated, confirming theoretical results through simulations and real data application.

Conclusions:

• The novel score tests, PM1 and PM2, offer substantial computational advantages for GWASs by efficiently managing missing genotype data.
• These methods maintain statistical power while drastically reducing computation time, making them valuable for large-scale genetic studies.
• The developed score tests are general and can be extended to other regression models.