Detecting gene-gene interactions using a permutation-based random forest method

Table 2 SNP interactions identified by permuted random forest (pRF)

SNP pairs	E1	E2	ΔE	MDR ranking
XPD.751, XPD.312	41.00 %	33.76 %	7.23 %	1
XRCC3, XPD.751	41.87 %	40.89 %	0.99 %	6
APE1, XPD.312	40.99 %	40.01 %	0.97 %	14
XRCC3, XRCC1.399	34.76 %	33.83 %	0.93 %	2
XRCC3, APE1	35.27 %	34.43 %	0.84 %	7
XPD.312, XRCC1.194	40.14 %	39.53 %	0.61 %	19
XPD.751, XRCC1.194	40.84 %	40.32 %	0.53 %	9
XRCC3, XPD.312	42.22 %	41.73 %	0.49 %	11
XPD.751, XRCC1.399	41.74 %	41.30 %	0.44 %	4
XRCC3, XRCC1.194	35.64 %	35.24 %	0.39 %	18
XRCC1.399, XPD.312	40.35 %	40.25 %	0.10 %	13
APE1, XPD.751	41.14 %	41.05 %	0.09 %	15
XRCC1.399, XRCC1.194	33.63 %	33.63 %	0.00 %	3
APE1, XPC.PAT	35.36 %	35.55 %	−0.19 %	17
APE1, XRCC1.194	34.04 %	34.22 %	−0.19 %	21
XRCC1.194, XPC.PAT	33.38 %	33.65 %	−0.27 %	20
XRCC3, XPC.PAT	35.00 %	35.37 %	−0.37 %	10
XRCC1.399, XPC.PAT	33.01 %	33.43 %	−0.42 %	12
XPD.312, XPC.PAT	40.48 %	40.95 %	−0.47 %	5
XPD.751, XPC.PAT	40.25 %	40.78 %	−0.53 %	16
APE1, XRCC1.399	33.86 %	34.71 %	−0.84 %	8

Classification error rates from datasets obtained by two permutation schemes, E1 (with main effects only) and E2 (with main effects and interaction), are shown in the table. Error rate differences were calculated and listed as Δ E. SNP pairs were ranked by their error rates differences in percentage, indicating the strength of interactions. SNP pairs column shows the permuted SNP names. MDR ranking column shows the ranking of top 2-way models MDR identified according to the results from our approach

ISSN: 1756-0381