Experimental Design SBC361

@yannick__ http://yannick.poulet.org

QMPlusFail

Programming in Rtest

Why consider experimental design?• If you’re performing experiments

• Cost • Time

• for experiment • for analysis

• Ethics • If you’re deciding to fund? to buy? to approve? to compete?

• are the results real? • allow clear interpretation? • can you trust the data?

Main potential problems

• Insufficient data/power

• Pseudoreplication

• Confounding factors

• Inappropriate statistics

Wrong Inaccurate & Misleading

Inappropriate designInappropriate implementationInappropriate analysis

(inappropriate interpretation)

Example: deer parasites• Do red deer that feed in woodland have more parasites than

deer that feed on moorland?

• Find a woodland + a moorland with deer; collect faecal samples from 20 deer in each.

• Conclusion?

• But: • pseudoreplication: (n = 1 not 20!):

• shared environment (influence each other) • relatedness

• many confounding factors: (e.g. altitude...)

Your turn: small & big Pheidole

workers.

• Is there a genetic predisposition for becoming a larger worker?

• Design an experiment alone.

• Exchange ideas with your neighbor.

e.g.: John.

Your turn again: protein production• Large amounts of potential superdrug takeItEasyProtein™

required for Phase II trials. • 10 cell lines can produce takeItEasyProtein™. • You have 5 possible growth media. • Optimization question: Which combination of temperature, cell

line, and growth medium will perform best? • Constraints:

• each assay takes 4 days. • access to 2 incubators (each can contain 1-100 growth tubes). • large scale production starts in 2 weeks

• Design an experiment alone. • Exchange ideas with your neighbor.

Taking measurements• How do you calibrate measuring instruments

(including human observers)? • Steps to reduce:

• subjective decision making? • inter-observer variability? • intra-observer variability? • Unusable/illegible measurements/notes

• Automation? • Avoid floor & ceiling effects • Ensuring that subjects are in “natural” conditions

do all that you can to ensure your design is robust

Overall

• Avoid easy mistakes

• Design & statistics are closely interlinked

• Consider biology carefully

• Better to spend more time planning.