A Filtering Approach to MDR Analysis

We have previously developed and evaluated a Tuned ReliefF (TuRF) method for filtering or reducing the number of SNPs in a genome-wide association study that are exhaustively analyzed using MDR. My previous post on TuRF can be found here. A pdf of the paper can be found here or can obtained by emailing me. Note that TuRF is available as part of the open-source MDR software package available from http://www.epistasis.org/.

TuRF was originally explored as an approach that could be applied to genome-wide association study (GWAS) data. However, we have more recently shown that TuRF can be equally useful for small candidate gene studies where perhaps only 20 or 30 SNPs have been measured. We have shown that reducing 20 or 30 SNPs to perhaps 5 with TuRF improved MDR analysis results. TuRF seems to be reducing the false-discovery rate by removing the noisy SNPs from the dataset prior to MDR analysis. This pre-processing step yields less chance for overfitting by MDR. We have observed in real datasets that running TuRF first reduces the training accuracy of the best model while increasing the testing accuracy. This suggests that MDR models of TuRF-selected SNPs are a better fit to the data. The following applied paper illustrating this point was just accepted for publication in Arthritis Research and Care.

Lorenzo Beretta, Francesca Cappiello, Jason H. Moore, Morena Barili, Casey S. Greene, Raffaella Scorza. Epistatic interactions of cytokine single nucleotide polymorphisms predict susceptibility to disease subsets in systemic sclerosis patients. Arthritis Research and Care, in press (2008).

Objective. Gene-gene interaction or epistasis is considered a ubiquitous component of complex human diseases such as systemic sclerosis (SSc). Epistasis is difficult to model by traditional statistical approaches, hence nonparametric computational algorithms, such as multifactor dimensionality reduction (MDR), have been developed. Methods. Two-hundred-forty-two consecutive unrelated Italian SSc patients and an equal number of well-matched healthy controls, were genotyped for 22 cytokine SNPs (13 cytokine genes). The distribution of the SNPs between controls and SSc patients and between controls and lcSSc or dcSSc patients was tested by the MDR constructive induction algorithm and by the logistic regression-based approach “focused interaction testing framework (FITF)”. Results. None of the studied SNPs had main independent effects on SSc or disease subset susceptibility and hence no epistatic interaction was detectable by FITF. The MDR analysis showed a significant epistatic interaction among the IL-2 G-330T, the IL-6 C-174G and the IFNg AUTR5644T SNPs and the IL-1R Cpst1970T, the IL-6 Ant565G and the IL-10 C-819T SNPs in lcSSc or dcSSc susceptibility, respectively. The relevance of the single multilocus attributes constructed by the MDR inductive algorithm was then confirmed by the parametric approach (p<0.001 for both control vs lcSSc and control vs dcSSc). Conclusion. We provide evidence for gene-gene interaction among cytokine SNPs in the context of SSc. The interaction among cytokine SNPs with a pro-fibrotic or a regulatory function on pro-fibrotic interleukins is relevant to the susceptibility to SSc subset and it appears to be more important than the contribution of any single cytokine SNP.