Table 6.
Estimates for the FN errors of each filtered dataset
| Dataset (ref.) | N | n | PFN | E [W1] | E [W2] | U obs | R obs |
| ItoFull [1] | 720 | 1,200 | 0.76 | 438 | 693 | 435 | 259,132 |
| ItoCore [1] | 128 | 100 | 0.38 | 47 | 14 | 43 | 8,156 |
| Uetz et al. [6] | 108 | 78 | 0.65 | 35 | 33 | 36 | 5,822 |
| Gavin et al. [10] | 852 | 2,429 | 0.44 | 1197 | 470 | 1,201 | 362,209 |
| Krogan et al. [11] | 1,458 | 11,744 | 0.80 | 3758 | 7,516 | 3,945 | 1,062,344 |
The expected number of false-negative (FN) errors on the filtered datasets for [1,6,10,11]. N is the number of proteins within each filtered dataset. The values for PFN and n are estimated upper bounds obtained by setting PFP = 0 and using the solution curves of Figure 6c,d. Denote W1 as the random variable for the number of unreciprocated FN observations, and W2 for the number of reciprocated FN observations. The variables Uobs and Robs show the observed number of unreciprocated and reciprocated interactions from the data, respectively. The table implies that in the worst case scenario for PFN, the doubly tested, reciprocated noninteracting protein pairs do not give us a conclusive indication about the presence or absence of an interaction. For this, more data are needed.