Determination of sample size for a multi-class classifier based on single-nucleotide polymorphisms: a volume under...

Area of Science:

• Genetics
• Bioinformatics
• Statistical Genetics

Background:

• Single-nucleotide polymorphisms (SNPs) are valuable for predicting phenotypes and disease risk.
• Limited clinical samples in multi-class classification necessitate efficient sample size determination.
• Area Under the ROC Curve (AUC) and Volume Under the ROC hyper-Surface (VUS) assess classifier performance.

Purpose of the Study:

• To develop a sample size determination method for SNP-based multi-class classification.
• To ensure cost-effectiveness and achieve pre-specified classification accuracy.
• To provide tools for scientists to assess sample adequacy.

Main Methods:

• Derived optimal Bayes and linear classifiers for coded SNP data (D ≥ 2 classes).
• Obtained normal approximations for classifier probabilities of correct classification.
• Developed a sample size determination method based on AUC/VUS differences, validated by Monte Carlo simulations.

Main Results:

• Validated the accuracy of AUC/VUS approximations using simulations.
• Demonstrated the sample size determination method's performance on HapMap data (3-4 populations, 92 SNPs).
• Determined required sample sizes for various population separation levels and threshold values.

Conclusions:

• Developed and illustrated a methodology for sample size determination in multi-class SNP classification.
• The methodology aids in assessing sample adequacy for achieving desired accuracy.
• An R package 'SampleSizeSNP' is available for practical application.