lab animal science Flashcards
Why talk about experimental design and
statistics?
- what are they
- which are the key steps
• What is science all about?
- Measuring stuff and demonstrating causes and
effects (cause → effect)
• What are experimental design & statistics all
about?
- Control of variability
- Isolating causal factors
- Demonstrating interactions between factors
• Which are the key steps in experimental design?
- Start out with a well-defined hypothesis and
research question
- Design the appropriate control group(s)
- Determine appropriate sample size.
- Randomly allocate experimental units to
treatments
- Control and reduce variability through blocking
(stratification) and factorial experiments
statistical power
- sample size, variability, and significance level affect
low power of a statistical analysis
- reduced chance of detecting a true effect
- low power reduces the likelihood that a statistically significant result reflects a true effect
Which factors determine whether our statistical test gives a
significant outcome?
B beta (); or Type-2 error (false negative) probability. (caused by biological variation) Power is calculated as (1-). Typically: 0.10 ≤ ≤ 0.20 E effect size (ES) and effect direction; one- or two-sided testing To be stated by the researcher A alpha (); or Type-1 error (false positive) probability (caused by biological variation) Conventionally: 0.01 ≤ ≤ 0.05 N n; number of experimental units (“number of animals”) To be calculated by the researcher S s; variability in measurements (standard deviation (s, ) or standard error (SE)) To be controlled by the researcher 41 NOTA BENEEE : Any effect size becomes statistically significant...when the sample size is large enough
How can you reduce variability?
Reducing variability by clever experimental design - fully randomized design (Full randomization doesn’t always work well with small
sample sizes as it can lead to segregation of treatments
or factors.)
Randomization and stratification
• If you can (and want to), fix a variable.
- e.g., use only 8-week old male mice from a single
strain.
• If you don’t fix a variable, stratify (block) it.
- e.g., use both 8-week and 12-week old male mice,
and stratify with respect to age.
• If you can neither fix nor stratify a variable,
randomize it.
B·E·A·N·S
̶ Knowing 4 parameter values, you can
calculate the 5th.
60
B – β, Type-2 error probability (1 - β = power)
E – Effect size (ES)
A – α, Type-1 error probability (significance
level)
N – Sample size (n)
S – Standard deviation (measure of
variability)
The bottom line when it comes to sample size
Sample size is proportional to the “signal-to-noise
ratio” in your population
So, sample size n can be reduced by: - Reducing noise, random variation () - Increasing the effect size (ES) • Reduction of noise through: - Standardization of protocols and measurements - Use of inbred strains - Clever experimental design (e.g. paired observations) 61 This illustrates an inverse square relationship between effect size and sample size: “To detect an effect 3 as small, sample size has to increase 32 = 9-fold.” (All else being equal.)
Size does matter! • A sample too small... - ...is difficult to replicate - ...is likely to produce a positive result that is a fluke - ...is a waste of time, money, means, and animals • Conclusions based on large samples are more reliable than those based on small samples. • But... a sample too large... - ...is also a waste of time, money, means and animals • Remember the 3 Rs: Replacement, Reduction, and Refinement. 62 Remember the inverse square relationship between effect size and sample size: “With a sample size 4 as large, you will be able to detect an effect only 4 = 2× as large.” (All else being equal.)
What are good effect sizes or effect size indices?
- Raw difference (delta) between two groups.
- Cohen’s d standardized difference.
- Pearson’s correlation coefficient, r.
- Odds ratio, OR.
- Risk ratio or relative risk, RR.
- Proportion explained variance, (ANOVA)
randomization
- why is it used
- methods
- what can be randomised
- to prevent bias (1) correct selection of experimental units (2) randomise experimental units to treatment (3) randomise order of measurements and animal housing (4) blinding and coding samples
- simple randomisation (paper or computer)
- latin square design (A Latin square is a block design with the arrangement of v Latin letters into a v×v array (a table with v rows and v columns). Latin square designs are often used in experiments where subjects are allocated treatments over a given time period where time is thought to have a major effect on the experimental response)
- blocked randomisation, stratification
- treatment groups, housing, order of treatment
Classification of animal models
Disease models
Induced (experimental) disease models (inducing obesity, tumor growth, renal damage, etc.)
Spontaneous (genetic) disease models (nude mice, hypertensive rat, etc.)
Transgenic disease models (KO, KI (i.e. introducing a crippled version of a GPCR), tissue selective, inducible)
Neutral models
Healthy animals from all species
…… inbred or outbred?
INBRED/OUTBRED
Inbred (isogenic) strains: produced by many (>20) generations of brother x sister mating. Aim:
Genetic homogeneity
The same genotype can be reproduced indefinitely, though over a period of time there may be some genetic drift due to the accumulation of new mutations.
Outbred strains: breeding colonies in which there is (certain) degree of genetic variation. Aim:
Genetically undefined outbred stocks
The amount of genetic variation depends on the breeding history of the particular colony.
inbreds will have a higher statistical reproducibility’ - Someone using an outbred stock generallyknows nothing about the genetic characteristics of individual animals, what genes they carry or how heterozygous they are. Background data on characteristics will be unreliable because they can change rapidly.
More than one strain can be used in a factorial experimental design without increasing the total number of animals
Why use INBRED
On average isogenic strains are more sensitive than outbred stocks to experimental treatments, which also increases the power of experiments which use them.
They are internationally distributed, so that work can be replicated all over the world.
Searchable lists of inbred strains of mice and rats and their characteristics are maintained by, for instance,
The fundamental principle of the controlled experiment is that treated and control groups should be identical, except for the treatment
validity of model
MODEL VALIDATION:
Best possible resemblance between model and research target
face validity
(resemblance in symptoms and signs)
predictive validity defined as the measure of how well a model can be used to predict currently unknown aspects of the disease in humans
(resemblance in reaction to current treatment - the similarity of the relation between, on the one hand, the triggering factors and the occurrence of the disease and, on the other hand, between the therapeutic agent and the disease)
construct validity
(resemblance in origin/mechanism)
what is genetic drift, vulnerability and what to do to prevent gd
constant tendency of genes t evolve even in the absence of selective forces. GD is fueled by spontaneous neural mutations that disappear or become fixed in a population at random
- single base changes, deletions, duplications, inversions in the DNA
- small colonies are more vulnerable to fix a mutation - colonies separated by 20 or more generations - phenotypic or genetic differences are discovered
- prevention: maintain pedigrees lines and detailed colony records
- watch for phenotypic changes in mutants and controls
- refresh breedes frequently (10 generations)
- avoid selection pressure
- verify genetic background with genome scanning
- cryopreserve unique strains
substrains? genetically different from parent colony
infections affect 3 things
quality of research, animal welfare and zoonoses
