Extension of multifactor dimensionality reduction for identifying multilocus effects in the GAW14 simulated data

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Multifactor dimensionality reduction (MDR) is a model-free method for detecting gene-gene and gene-environment interactions in case-control studies.
• Existing MDR methods often rely on cross-validation, which can be computationally intensive.

Purpose of the Study:

• To explore modifications of the MDR method, creating an extended MDR (EMDR).
• To evaluate EMDR's ability to identify genetic effects using simulated data.
• To compare different statistical approaches within EMDR for model selection and significance testing.

Main Methods:

• Developed EMDR with model selection not requiring cross-validation.
• Utilized a chi-square statistic as an alternative to prediction error (PE).
• Implemented three permutation tests (fixed, non-fixed, omnibus) for p-value determination of multilocus models.

Main Results:

• Chi-square and PE statistics yielded consistent results.
• EMDR without cross-validation produced results comparable to 10-fold cross-validation.
• Non-fixed and omnibus permutation tests effectively controlled false positives, unlike the fixed test; the omnibus test was most conservative.

Conclusions:

• EMDR without cross-validation provides accurate results efficiently.
• The non-fixed permutation test offers a favorable balance between statistical power and false-positive rates.
• EMDR enhancements improve the identification of genetic effects in complex trait studies.