Data requirements
This section describes the requirements used to process the IMPC data. It covers mutant sample sizes and minimum animal numbers, specimen requirements for statistical analysis, and analyses outside the standard statistical pipeline, including the treatment of viability, embryo gross pathology, histopathology, pathology, fertility, and other specialised datasets.
Mutant sample sizes and minimum animal numbers
To detect lethality, the IMPC Viability Primary Screen screens at least 28 pups. The target number of adult knockout animals to phenotype is 14: 7 females and 7 males. The n-number criteria for statistical analysis is determined independently of the experimental n-numbers and vary depending on the data type, please refer to the “Specimen Requirements for Statistical Analysis” section.
IMPReSS sets a minimum number of specimens for each procedure and these are specified in the experimental design section of the procedures on IMPReSS.
The number of animals reaching the statistical pipeline may be fewer than the numbers specified on IMPReSS as some animals may not be able to go through the procedures, for example for welfare reasons, or data have been excluded for QC reasons. Usually for these cases a status code is submitted to justify the lack of data and this can be taken into account before the data goes through the statistical pipeline. In rare cases, such as with Eye Morphology and Combined SHIRPA and Dysmorphology data, the procedure set up requires a value even when no data can be obtained and the number of specimens analysed may be lower for these parameters, because categories with “No data” are excluded from the analysis.
Specimen requirements for statistical analysis
Procedure-level specimen requirements are distinct from the eligibility checks applied to individual parameters by the statistical pipeline. The number of observations available for analysis may be smaller because measurements can be missing, fail quality control or be excluded during preprocessing.
The n-number criteria for statistical analysis is determined independently of the experimental n-numbers and vary depending on the data type:
The statistical pipeline applies additional eligibility checks after preprocessing. Observations with missing response values are excluded, and both experimental and control observations must be present.
- Continuous analysis pipeline: Requires four data points per genotype-sex group, including controls, except for the Auditory Brain Stem Response screen, which requires only one data point per genotype-sex group. This requirement was established in consultation with biological and data domain experts. Additionally, for the Reference Range Plus method the analysis is only performed when the control count is more than 60 animals for each relevant sex to the experiment.
- Categorical analysis pipeline: Does not apply the continuous-data minimum, but requires both genotype groups, more than one mutant observation in at least one genotype-sex group, and compatible mutant and control representation within at least one sex.
Analyses are not performed when:
- Every usable measurement is identical; there is no difference to investigate.
- Almost all the results are the same: The pipeline has a second check that skips numeric data when there are too few different values compared with the number of measurements. This is a built-in cutoff intended to avoid analysing data with very little variation.
- The measurement or procedure is on the skip list: Some types of measurements or procedures are marked as unsuitable for analysis by this pipeline, so they are skipped automatically. For more information please refer to the analyses outside the standard statistical pipeline.
- No usable measurements remain: The pipeline cleans the data first, for example by removing missing results. If nothing is left to analyse afterward, it skips that group.
Analyses outside the standard statistical pipeline
Time-series parameters are not analysed by the statistical pipeline. Additional datasets are also treated separately with a custom approach: fertility, viability, embryo viability, embryo gross morphology, pathology, and histopathology. Experts assign the phenotypes using statistical methods suited to the breeding scheme at each phenotyping centre. IMPReSS provides specific information on how adult and embryo viability as well as fertility data are analysed.
Viability
- Version 1: The mp_term is assigned based on the outcome (homozygous viable or homozygous lethal).
- Version 2: The pipeline determines viability or lethality and assigns the mp_term based on the percentage of living and viable pups.
- The same process applies to embryo viability.
Embryo Gross Pathology
- If there are two or more specimens, the pipeline marks it as abnormal and significant. Specimens can be homozygous, hemizygous, or heterozygous.
Histopathology
- Experts provide a phenotype term with a significance score, 0 for not significant and 1 for significant. The CDA then identifies a significant association between the allele and the phenotype based on the technical expert’s opinion.
Gross Pathology
- Phenotypes are identified by CDA by comparing the proportion of abnormal KO specimens with the proportion of abnormal WT specimens for each colony and parameter. A result is considered significant when the KO proportion is at least 25 percentage points higher than the WT proportion, based on at least four KO specimens and 60 baseline animals for early adult data and 30 baseline animals for late adult data. If WT data are unavailable, at least 25% of KO specimens must be abnormal. A single abnormal KO specimen does not automatically make a result significant.
3i and Pain Projects
- In addition, analyses of data contributed by the pain working group and 3i project were assessed by external professionals and are not processed by the statistical analysis pipeline.