Performance of synthetic accessibility as compared to FBA, MOMA, EMA, and other topology-based measures using the *E. coli* metabolic network. The graphs illustrate the relative performance of the techniques using two measures, accuracy, (*TP* + *TN*)/(*TP* + *TN* + *FP* + *FN*), and the negative log of the *χ*^{2} statistic's *p*-value, which indicates the correlation between the in silico predictions and the in vivo observations of *E. coli* mutant strain viability. The *χ*^{2} statistic is calculated using a contingency table like the ones in for the smaller data set (79 data points, 90 data points for EMA), the insertional mutant data set (487 data points), and the combined data set (560 data points) (,,). When the larger, more representative insertional mutant data set or the combined data set is used, synthetic accessibility is as accurate and statistically significant as for FBA. However, synthetic accessibility performs more poorly on the smaller data set, probably because this data set has few data points and only covers central metabolism, a small fraction of the whole metabolic network. The other topology-based measures, degree and diameter, perform worse than FBA, MOMA, EMA, and synthetic accessibility, indicating that they poorly characterize the functioning of the metabolic network. The random predictions are made using the expected values produced for the FBA *χ*^{2} test and represent the expected performance if there were no correlation between the in silico and in vivo predictions. They vary very little if the expected values for the other *χ*^{2} tests are used.

## PubMed Commons