Table 1.
Maximum likelihood solution for the spoke model (ψ = 3.5) and the matrix model (ν = 10.0). We choose the number of clusters that maximizes the likelihood by searching over a range of values of K. The estimated the false negative rate is denoted by ν* and the estimated false positive rate by φ*. For comparison we show the error estimates based on the MIPS complexes, νMIPS and φMIPS, restricted to proteins with MIPS annotation. See also Table 2.
| Dataset | K | ν* | φ* | νMIPS | φMIPS | |
| Gavin02 | Spoke model | 393 | 0.423 | 1.3 × 10-3 | 0.598 | 6.5 × 10-3 |
| Matrix model | 310 | 0.752 | 1.7 × 10-3 | 0.717 | 5.2 × 10-3 | |
| Gavin06 | Spoke model | 698 | 0.547 | 2.4 × 10-3 | 0.637 | 8.3 × 10-3 |
| Matrix model | 550 | 0.807 | 2.7 × 10-3 | 0.901 | 6.4 × 10-3 |