FIG. 4.

Reproducibility analysis. (A) Reproducibility probability in 10 groups. The figure shows the probability that a TFG interaction found in overlap group x (horizontal axis) using 10 datasets is found in group x+2 when 12 datasets are used. This probability gives an estimate of reproducibility for group 1 to group 10. The 10 points were found to be well approximated by a single-parameter sigmoid function of the form 1/[1 + Exp(z – x)], where x is the overlap group and the fitting parameter z = 6.15609 (R² = 0.9986). (B) Reproducibility probability distribution within group 2. The figure shows the distribution of the probability of reproducibility (after the addition of two more datasets) as a function of the rank R of an interaction within group 2 with k = 5 to 10 datasets. The fitting to the probability distribution is of the form P(R,k) = a Exp[−R/Q(k)], where R is the rank of the interaction within the group, and Q(k) is a characteristic decay rank that depends on the number of datasets k. The factor a takes into account the probability distribution normalization. This figure indicates that the correlation-based ranking within group 2 contains less information and is less significant when more datasets are added, since the top-ranking interactions in this group become less reproducible.