Objective: We conducted a simultaneous analysis of candidate genetic loci for their genotypic association with the susceptibility to vascular dementia (VaD) to put forth the best model for predicting genetic susceptibility to VaD.
Methods: Individual-locus effects and their epistatic effects on susceptibility to VaD were simultaneously assessed by multifactor dimensionality reduction and entropy-based method. The 23 loci in 12 genes were studied in 207 VaD patients and age-matched and sex-matched 207 controls.
Results: The multifactor dimensionality reduction analysis revealed that the best single-locus candidate model included angiotensinogen (AGT) Thr235Met with testing accuracy (TA) of 58.31%, the best two-locus candidate model included AGT Thr235Met and transforming growth factor-beta1 Pro10Leu with TA of 58.06%, the best three-locus candidate model was not significant (P>0.05), and the best four-locus candidate model included transforming growth factor-beta1 Pro10Leu, AGT Thr235Met, sterol regulatory element binding protein 2 G34995T, and leukemia inhibitory factor T4524G with TA of 57.13% (P<0.05). The best four-locus model was, however, still in question because of the inconsistent best model selection by cross-validation. Synergistic epistatic effect of the best two-locus model was proven by entropy-based estimation.
Conclusion: The best predictor for genetic susceptibility to VaD was the single-locus model of AGT. The best two-locus model reflecting epistasis would be also employed for predicting its susceptibility. Further studies on the epistasis are to elucidate their underlying mechanisms.