Your SlideShare is downloading. ×
0
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Alec Zwart - Statistical experiment design principles for biological studies
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Alec Zwart - Statistical experiment design principles for biological studies

1,148

Published on

“To consult the statistician after an experiment is finished is often merely to ask them to …

“To consult the statistician after an experiment is finished is often merely to ask them to
conduct a post mortem examination. They can perhaps say what the experiment died of.” ‐ Sir Ronald Aylmer Fisher, the father of modern statistics.
Statistical experimental design, accompanied by the appropriate statistical analyses, plays a crucial role in producing valid and precise inferences, and avoiding ‘design disasters’ in empirical science. I will quickly refresh some of the basic elements of experimental design in general, and discuss some key issues and examples that arise
in the areas of genetics/genomics and high throughput data.

First presented at the 2014 Winter School in Mathematical and Computational Biology http://bioinformatics.org.au/ws14/program/

Published in: Science
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
1,148
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
15
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  1. Statistical experimental design  principles for biological studies Alexander (Alec) Zwart 10 July 2014 CSIRO DIGITAL PRODUCTIVITY AND SERVICES FLAGSHIP
  2. An apology, and a caveat • Apology: I am not a genetics/genomics expert! •But I DO know a bit about design… • Caveat: Experimental design is a science! (as is  statistical analysis…) •Every experiment is different •This talk is necessarily simplistic, and very incomplete… •Intended as a reminder of some concepts, some bits  of advice, and a plea… Presentation title  |  Presenter name2 |
  3. An experiment – recap: • Apply treatments (experimental conditions) or  treatment combinations to subjects (or experimental  units). • Carefully control/remove other influences as much as  possible. • Observe one or more responses (observed or measured  results) • A way to infer cause => effect. • Genotype/Species/Variety is a treatment?  Yes... Presentation title  |  Presenter name3 |
  4. What is statistical experimental design  concerned with? • Everything!  The entire process leading to the  final dataset… • Recognising sources of variation (systematic,  random) that influence your results and hence  controlling for them •How to ensure that we obtain data that  provides ‘honest’ and accurate answers to  our questions Presentation title  |  Presenter name4 |
  5. Key point: • Good statistics cannot rescue a bad dataset! • And a bad dataset may not be able to answer  your questions (honestly, that is) • Do the experiment right => get a good dataset • Consider design from the very beginning of  planning the experiment Presentation title  |  Presenter name5 |
  6. Fundamentals Randomisation Blocking Replication Presentation title  |  Presenter name6 |
  7. Randomisation • Allocate treatments to subjects randomly (ideally using a proper random number  generator, not your own personal judgement) • Also randomise: •Spatial layout •Order of processing • Avoid systematic patterns (But: see Blocking) •Protects against unforeseen biases ED for Biology |  Alexander Zwart7 |
  8. Blocking •Identify groupings that may effect your  results •Choice of operator/technician/machine •Cage/Table/shelf/assay plate/batch •Try to assign (ideally) one of each  treatment to each group! (Randomised  Block Design or RBD) •(Example coming later…) ED for Biology |  Alexander Zwart8 |
  9. Replication • Sample size •G*Power (http://www.gpower.hhu.de/en.html) •Russ Lenth’s applets  (http://homepage.stat.uiowa.edu/~rlenth/Power/) •Simulation studies in the literature? (Google scholar) • Biological/internal/technical replicates •Don’t confuse these – Internal/technical replicates are  not a substitute for biological replication. •(more later…) ED for Biology |  Alexander Zwart9 |
  10. 12 mice, three treatments (Colour) Presentation title  |  Presenter name10 | Completely randomised design Balanced (equal info on each treatment, treatments compared with equal accuracy)
  11. But, the mice must be housed… Presentation title  |  Presenter name11 | Here, treatments are confounded with cages Cannot separate Cage effects from Trt effects!
  12. Cages as Blocks Presentation title  |  Presenter name12 |
  13. Grr Presentation title  |  Presenter name13 | Animals (especially) can pose a problem… Grrr! Grrr! GrrEeeek!
  14. Cages as Experimental Units / Subjects… Presentation title  |  Presenter name14 | …& Cage = Mouse
  15. (Aside‐ env. trend across columns of cages? Presentation title  |  Presenter name15 | => Block by Columns!
  16. Not enough cages? A compromise: But, Cages are the Experimental units Only two replicates of each Trt!
  17. More on replication: • The replication used determines the population(s)  being sampled... ...hence… • The replication used determines the conclusions that  may be validly drawn from a statistical test ED for Biology |  Alexander Zwart17 |
  18. Comparison between populations of plants ED for Biology |  Alexander Zwart18 | Mutant 2.1 1.4 1.9 1.6 4.3 4.44.13.9 Wild Type
  19. Comparisons between two plants ED for Biology |  Alexander Zwart19 | Mutant 4.3 4.44.13.9 Wild Type 2.1 1.4 1.9 1.6
  20. Recording • Details of treatments and associated  randomisation, blocking, types of replication  should always be recorded • Check what information is required for the  analysis, or talk to the statistician about what is  required. • This is important meta data when distributing  your data… ED for Biology |  Alexander Zwart20 |
  21. More on Blocking… • ‘One of each treatment per block’ is not always  feasible – blocks may be too small •‘Incomplete blocks’ • Think about: •Balance – treatments occur together in the same  blocks the same number of times. •Linkage – common treatments in different blocks  provide links between blocks – ensure you have  plenty of linkage. ED for Biology |  Alexander Zwart21 |
  22. Balanced Incomplete Block Design (BIBD) - Trts occur together the same number of times - Plenty of treatment-linkages between blocks
  23. Notes on incomplete blocks • BIBDs are only availble for certain combinations  of parameters. • However, the general principles are still relevant •‘Unbalanced incomplete blocks’ • Note that teasing apart the treatment effects  from block effects in unbalanced situations is  dependent on the capabilities of the analysis  method •Check on said capabilities, consider simpler  experiment if necessary… ED for Biology |  Alexander Zwart23 |
  24. Treatment structure • Sir Ronald Fisher => factorial treatment structure (3x4 factorial) • 5 = Number of replicates of each treatment ED for Biology |  Alexander Zwart24 |
  25. Factorial treatment structure • Separate main effects and interactions • Can include more factors (in theory) • Unbalanced numbers of treatments usually OK: ED for Biology |  Alexander Zwart25 |
  26. But beware missing treatment combinations • The more that entire treatments are missing  from the factorial, the harder it is to tease apart  main effects and interactions and the more  confounded different treatment effects become. ED for Biology |  Alexander Zwart26 |
  27. Don’t get too ambitious! • Two‐way factorials are much easier to interpret  than three‐way, four way… etc. • The higher the factorial structure, the more  likely the structure will be incomplete  •(missing values, treatment combos which don’t make  sense, treatment combos you aren’t interested in and  don’t include). • Statisticians and their tools can’t work miracles! • Simpler is better… • Talk to a statistician! ED for Biology |  Alexander Zwart27 |
  28. Plea: •If you don’t already know how to design  and conduct good experiments… …then… •…involve a statistician in the early stages of  planning your experiment! •And note – the statistician may need time… Presentation title  |  Presenter name28 |
  29. Sir Ronald Fisher, ‘the father of modern  statistics’… “To consult the statistician after an experiment  is finished is often merely to ask them to  conduct a post mortem examination.  They can perhaps say what the  experiment died of.” (R. Fisher) Presentation title  |  Presenter name29 |
  30. CSIRO DIGITAL PRODUCTIVITY AND SERVICES FLAGSHIP Thank you

×