Table 3.
Top 10 most significant triplets identified by supervised MethReg analysis of ROSMAP AD dataset
| Triplet data | Annotation | DNAm group × TF activity | TF–target | AD literature | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Probe ID | Position | TF | Target gene | CpG–target distance | TF.role | DNAm.effect | Estimate | P-value | Adj. P-value | Correlation | P-value | TF | Target gene |
| Promoter analysis | |||||||||||||
| cg17418085 | chr1: 31229122 | SPI1 | LAPTM5 | 1543 | Dual | Invert | −7.177 | 7.36E−12 | 3.06E−07 | −0.011 | 8.05E−01 | (105,106) | (109) |
| Distance-based analysis | |||||||||||||
| cg11556846 | chr2: 200468728 | ESR1 | SATB2 | −132 738 | Dual | Invert | −7.266 | 3.03E−07 | 7.80E−03 | −0.058 | 1.82E−01 | (124,126) | (157) (158) |
| cg00153919 | chr16: 88859944 | CEBPD | PIEZO1 | −8324 | Repressor | Enhancing | −8.122 | 4.92E−07 | 9.14E−03 | −0.044 | 3.10E−01 | (130,131) | (159) |
| cg08824847 | chr11: 35052388 | NR3C1 | PDHX | 115 011 | Activator | Enhancing | 4.383 | 6.13E−07 | 1.02E−02 | 0.06 | 1.66E−01 | (114,116) | (117,118) |
| cg08760493 | chr4: 109994039 | TCF12 | SEC24B | −360 887 | Dual | Invert | 8.120 | 1.26E−06 | 1.49E−02 | 0.033 | 4.44E−01 | (132,133) | (160) |
| cg05715492 | chr7: 98991138 | SRF | ARPC1B | 19 265 | Repressor | Attenuating | 11.581 | 1.34E−06 | 1.49E−02 | 0.017 | 7.00E−01 | (135) | (161) |
| cg09316954 | chr16: 67687754 | TCF12 | CARMIL2 | 8931 | Repressor | Attenuating | 8.376 | 2.28E−06 | 2.24E−02 | −0.07 | 1.08E−01 | (132,133) | (162) |
| cg21155834 | chr2: 149282209 | GABPA | ORC4 | −503 061 | Activator | Enhancing | 6.407 | 2.75E−06 | 2.41E−02 | 0.045 | 2.98E−01 | (137) | (163) |
| cg13819552 | chr9: 95799870 | TCF12 | ZNF484 | −159 565 | Activator | Enhancing | 7.989 | 2.92E−06 | 2.41E−02 | 0.039 | 3.71E−01 | (132,133) | (164) |
| cg21535772 | chr2: 171679906 | NFE2L2 | AC007405.4 | 52 282 | Dual | Invert | 7.065 | 3.10E−06 | 2.41E−02 | −0.026 | 5.47E−01 | (138,165) | (166) |
Triplet data included the CpG located in promoter regions (<2 kb from TSS) or distal regions (>2 kb from TSS), its genomic location (hg19 genome), the target gene and the TF with binding sites within 250 bp of the CpG. Annotations included the distance between CpG and target gene (CpG–target distance), MethReg predicted roles of the TF on the target gene and the effect of DNAm on the TF. TF activity (hereafter referred to as TF) was first estimated using the GSVA software. After removing age at death, sex, batch and cell type effects in gene expression data and methylation data separately, the robust linear model target gene expression residual ∼ DNAm group + TF + TF × DNAm group was then fitted to data to obtain estimated TF × DNAm group effect (estimate), its P-value and FDR (adj. P-value) for the term DNAm.group × TF. TF–target association in all samples was estimated using a correlation coefficient.