Statistics on big biomedical data: Methods and pitfalls when analyzing high-throughput screens
Download as PPT, PDF1 like89 views
This document discusses statistical methods for analyzing high-throughput biomedical screens and common pitfalls. It introduces several statistical tests such as t-tests, ANOVA, Fisher's exact test, and the Mann-Whitney U test. It also discusses challenges like multiple testing, resampling techniques, and biases that can occur like studiedness bias and abundance bias in big data analyses. Controlling false discovery rates and considering effect sizes are recommended over solely relying on p-values to determine biological significance.
Statistics on big biomedical data: Methods and pitfalls when analyzing high-throughput screens
- 1. Statistics on big biomedical data Methods and pitfalls when analyzing high-throughput screens Lars Juhl Jensen
- 2. Statistics on big biomedical data Methods and pitfalls when analyzing high-throughput screens Lars Juhl Jensen
- 3. t-test
- 4. ANOVA
- 6. useful tests
- 7. counts
- 9. Jensen et al., Nature Reviews Genetics, 2012
- 11. real numbers
- 14. do the medians differ?
- 16. medians can mislead you
- 17. do the distributions differ?
- 20. does not tell how they differ
- 21. resampling
- 26. multiple testing
- 27. xkcd.com
- 28. xkcd.com
- 29. xkcd.com
- 30. xkcd.com
- 33. Bonferroni
- 34. avoid making any errors
- 35. too conservative
- 37. control false discovery rate
- 39. resampling
- 40. negative set
- 42. Huang et al., Journal of Proteome Research, 2014
- 43. studiedness bias
- 44. we study disease proteins
- 45. thus we know many PTMs
- 46. abundance bias
- 47. higher expressed
- 48. easier to detect in assays
- 51. the big data effect
- 52. if you have enough data
- 53. any difference is significant
- 54. but maybe not relevant
- 56. “significant”
- 58. p-value
- 60. effect size
- 62. volcano plots
- 63. Lundby et al., Science Signaling, 2013
- 64. rather ad hoc
- 65. questions?