Table 2.
Summary of the results from each of the methods used to derive maternal and offspring specific genetic effects using summary statistics from a GWAS of own birth weight and a GWAS of offspring birth weight.
| Method | No sample overlap | Sample overlap | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Comparison with SEM using individual level data | Computational time (min)a | LD score regression intercept (standard error) | Number of SNPs with P < 5 × 10−8 (Number of loci)b [Number of false positive loci]c | Comparison with SEM using individual level data | Computational time (min)a | LD score regression intercept (standard error) | Number of SNPs with P < 5 × 10−8 (Number of loci)b [Number of false positive loci]c | ||||
| Effect estimate | Standard error | Effect estimate | Standard error | ||||||||
| SEM using summary statisticsd | Offspring effect | Accurately estimated | Deflated | 4641 | 1.754 (0.012) | 2143 (546) [486] | Accurately estimated | Comparable | 5753 | 1.066 (0.009) | 792 (38) [19] |
| Maternal effect | Accurately estimated | Deflated | 1.197 (0.009) | 728 (76) [51] | Accurately estimated | Comparable | 1.058 (0.008) | 656 (43) [18] | |||
| Linear approximation of SEM | Offspring effect | Accurately estimated | Comparable | 42 | 1.012 (0.007) | 37 (6) [1] | Accurately estimated | Inflated | 34 | 0.939 (0.008) | 320 (13) [2] |
| Maternal effect | Accurately estimated | Comparable | 1.016 (0.007) | 423 (18) [0] | Accurately estimated | Inflated | 0.941 (0.007) | 496 (16) [0] | |||
| MTAG | Offspring effect | No correlation with effect estimate from SEM | Deflated | 30 | 0.988 (0.011) | 3595 (80) [5] | No correlation with effect estimate from SEM | Deflated | 30 | 1.001 (0.012) | 5461 (128) [16] |
| Maternal effect | No correlation with effect estimate from SEM | Deflated | 1.005 (0.011) | 3880 (89) [11] | No correlation with effect estimate from SEM | Deflated | 1.005 (0.012) | 5351 (119) [18] | |||
| mtCOJO | Offspring effect | Consistently underestimated | Deflated | 48 | 1.022 (0.007) | 62 (7) [1] | Consistently underestimated | Deflated | 62 | 1.049 (0.009) | 896 (22) [2] |
| Maternal effect | Consistently underestimated | Deflated | 36 | 1.031 (0.008) | 386 (17)[0] | Consistently underestimated | Deflated | 63 | 1.043 (0.008) | 524 (18) [0] | |
| Genomic SEMd | Offspring effect | Accurately estimated | Comparable | 103 | 0.985 (0.007) | 32 (16) [1] | Accurately estimated | Comparable | 224 | 0.971 (0.008) | 394 (13) [2] |
| Maternal effect | Accurately estimated | Comparable | 0.978 (0.007) | 372 (16) [0] | Accurately estimated | Comparable | 0.976 (0.007) | 532 (18) [0] | |||
Computational time is used as a guide only to compare between methods; this will differ depending on the computing resources available for each analysis.
aAnalyses were conducted on a Dell Poweredge R840 server that is part of a Rocks 7 open-source Linux cluster based upon CentOS 7.4. Specific details include: CPU: 4 × Intel Xeon Gold 5117 2.0 G, 14 C/28 T, 10.4GT/s, 19.25 M Cache, Turbo, HT (105 W); Disk: 480GB SSD SATA Read Intensive 6 Gbps 512 2.5in; Memory: 24 × 64GB LRDIMM, 2666MT/s, Quad Rank. The number of cores and memory were assigned to each job to optimize performance of the method.
bA locus was defined as 500 kb from the sentinel SNP.
cA false positive was defined as a locus that is greater than 500 kb from the already known birth weight associated sentinel SNP.
dWe ran the 22 chromosomes in parallel for the SEM using summary statistics and Genomic SEM, so the computational time was the time to run chromosome 2 as this is the longest chromosome and therefore slowest to complete.