Points of Significance: Replication

Journal name:
Nature Methods
Volume:
11,
Pages:
879–880
Year published:
DOI:
doi:10.1038/nmeth.3091
Published online

Quality is often more important than quantity.

At a glance

Figures

  1. Replicates do not contribute equally and independently to the measured variability, which can often underestimate the total variability in the system.
    Figure 1: Replicates do not contribute equally and independently to the measured variability, which can often underestimate the total variability in the system.

    (a) Three levels of replication (two biological, one technical) with animal, cell and measurement replicates normally distributed with a mean across animals of 10 and ratio of variances 1:2:0.5. Solid green (biological) and blue (technical) dots show how a measurement of the expression (X = 12) samples from all three sources of variation. Distribution s.d. is shown as horizontal lines. (b) Expression variance, Var(X), and variance of expression mean, Var( ), computed across 10,000 simulations of nAnCnM = 48 measurements for unique combinations of the number of animals (nA = 1 to 48), cells per animal (nC = 1 to 48) and technical replicate measurements per cell (nM = 1 and 3). The ratio of Var(X) and Var( ) is the effective sample size, n, which corresponds to the equivalent number of statistically independent measurements. Horizontal dashed lines correspond to biological and total variation. Error bars on Var(X) show s.d. from the 10,000 simulated samples (nM = 1).

  2. The number of replicates affects FDR and power of inferences on the difference in variances and means.
    Figure 2: The number of replicates affects FDR and power of inferences on the difference in variances and means.

    Shown are power and FDR profiles of a test of difference in cell variances (left) and animal means (right) for 48 (nM = 1) or 144 (nM = 3) measurements using different combinations of nA and nC. Vertical arrows indicate change in FDR and power when technical replicates are replaced by biological replicates, as shown by nA,nC,nM, for the same number of measurements (144). Values generated from 10,000 simulations of a 10% chance of a treatment effect that increases cell variance 2σC2 or animal mean 1.1 × μA. Samples were tested with two-sample t-test (sample size nA) at two-tailed α = 0.05.

References

  1. Robasky, K., Lewis, N.E. & Church, G.M. Nat. Rev. Genet. 15, 5662 (2014).

Download references

Author information

Affiliations

  1. Paul Blainey is an Assistant Professor of Biological Engineering at MIT and Core Member of the Broad Institute.

  2. Martin Krzywinski is a staff scientist at Canada's Michael Smith Genome Sciences Centre.

  3. Naomi Altman is a Professor of Statistics at The Pennsylvania State University.

Competing financial interests

The authors declare no competing financial interests.

Author details

Additional data