As expected, most biological replicates are strikingly similar to one another (Additional file 2: Figure S1B-F).
However, due to the high cost, recent studies rarely used biological replicates; we therefore focus on the situations where at most technical replicates are used and Poisson distribution is assumed to be appropriate.
As a reference, we performed an unsupervised hierarchical clustering including all technical replicates (Additional file 8), which showed that phosphorylation of peptides by cell lysates of most technical replicates was comparable.
Time points where the replicates are most dissimilar contribute least to the likelihood as ϵ t is larger as is desirable.
Although technical replicates can improve the precision and the reliability of the measurement and provide information for quality control, biological replicates are most effective in reducing the variance of the estimate of mean difference.
Unfortunately, WGBS is still prohibitively expensive for most laboratories, particularly when many replicates are required; significantly, a high fraction of reads contain no methylated cytosines.
(Compare Bunnik Figure 1 A-D) Finally, sequencing coverage and biological replicates are the most important determinant of obtaining reliable data in our experiments.
In general, this is a difficult task, as for most experiments only very few replicates are available; for example the ENCODE consortium set a standard of two independent biological replicates per ChIP-Seq measurement [ 25].
This is because in the Wright Fisher model, the realized variance among individuals in genes contributed to the next generation (V k *) is a random variable; effective size equals N only when V k * is exactly equal to the binomial expectation E(V k ) = 2(N−1)/ N, so in most replicates Ne* is higher or lower than N because V k * ≠ 2(N−1)/ N.
The average profiles of most replicate pairs are highly consistent.
