Table 1.
Description of workflow tested.
Figure 1.
Shown are chloride blood chemistry measures collected for Expitm1a(KOMP)Wtsi/Expitm1a(KOMP)Wtsi (HOM) and wildtype (+/+) mice at week 16 with the high throughput pipeline. The phenotype was classed as female specific effect (p value: 0.0016, Genotype by female effect quantified as −5.34±1.29 (se) with PhenStat MM without weight). Comparison based on 298 female and 264 male wildtype mice and 7 female and 6 male knockout mice.
Figure 2.
Variation of chloride readings with time.
Shown are chloride blood chemistry measures collected at week 16 by batch for those collected within 2012 for both knockout and control mice collected on the WTSI MGP Select high throughput pipeline for the core strain B6N. The dotted lines indicate the 95% percentile values. The boxplot highlighted in black shows the date on which the Expitm1a(KOMP)Wtsi female data were collected, and shows that all values collected on that date were low. Whilst these data points were low, the instruments daily QC checks were within the required boundaries.
Figure 3.
Example temporal variation in control data.
Shown are box and whisker plots as function of batch (x-axis) A: Fat mass readings for Institut Clinique de la Souris IMPC pipeline control C57BL/6NTac male mice B: Platelet readings for the German Mouse clinic IMPC pipeline control C57BL/6NTac(USA) male mice.
Figure 4.
Variation in distribution of p-values with workflow.
Example empirical distribution profiles for the resampling of the seven traits measured in the Dual-energy X-ray absorptiometry control data. A and B: Multi-Group workflow where A shows the test of genotype effect and B the test of genotype-by-sex effect. C and D: One batch per colony workflow where C shows the test of genotype effect and D the test of genotype-by-sex effect.
Figure 5.
False Positive Rates for resampling control data for various workflows.
Shown are FPRs under the null hypothesis of no phenotypic effect, estimated by resampling controls for various workflows, for 70 traits from five assays from the WTSI Mouse Genetics Project (MGP) Select Pipeline. The y-axis shows box-and-whisker plots of the distribution of the FPR, defined as the fraction of resampled datasets significant at the nominal 5% level in a mixed model. A: FPR of the test of genotype effect. B: FPR of the test of sex by genotype interaction. The labels relate to the workflows as defined in table one where MG indicates the Multi-Group, R the Random, B2 the TwoBatch, B1 the OneBatch and B3 the ThreeBatch workflows.
Figure 6.
False Positive Rates for other institutes for various workflows.
Shown are FPRs under the null hypothesis of no phenotypic effect, estimated by resampling controls from two independent institutes for various workflows. The y-axis’s show box-and-whisker plots of the distribution of the FPR, defined as the fraction of resampled datasets significant at the nominal 5% level in a mixed model. A, B: Resampling results using data from the German Mouse Clinic, where A is the FPR for genotype effect and B is the FPR for genotype-by-sex effect. C, D: Results using control data from Institut Clinique de la Souris, where C is the FPRs for the genotype effect and D the FPRs for the genotype-by-sex effect. The labels relate to the workflows as defined in table one where MG indicates the Multi-Group, R the Random, B2 the TwoBatch, B1 the OneBatch and B3 the ThreeBatch workflows.
Figure 7.
False Positive Rates for simulated data for various workflows.
Shown are FPRs under the null hypothesis of no phenotypic effect, estimated by resampling simulated controls. The y-axis’s show box-and-whisker plots of the distribution of the FPR, defined as the fraction of resampled datasets significant at the nominal 5% level in a mixed model. A: FPRs of the test of genotype effect. B: FPRs of the test of genotype-by-sex effect. The labels relate to the workflows as defined in table one where MG indicates the Multi-Group, B2 the TwoBatch, B1 the OneBatch and B3 the ThreeBatch workflows.
Figure 8.
Power analysis - impact of workflow and batch variation.
The mixed model methodology sensitivity was assessed with a resampling study on simulated control data where a signal as a proportion of the biological variation was added to construct a knockout group prior to statistical comparison. A: Impact of increasing batch noise, where the noise was a multiplier of 10% of the biological noise. Data shown are the output for simulated bone mineral density trait with a random workflow and a 0.05 significance threshold. B: Impact of workflow. Data shown are the output for the five workflows considered within this manuscript for a 0.05 significance threshold.