Figure 7

Area under curve of precision-recall (AUCPR) of GTRNetwork algorithm combinations with different input TF-gene network topologies. The performance of GTRNetwork is relatively consistent while using input TF-gene network topologies containing different percentages of known regulatory links, except using the 90% of known regulatory links as the input TF-gene network topology. EM-based or SVD-based TFA prediction algorithms (E/S-C-C, E/S-C-N, E/S-A-C, E/S-A-N) give significantly better performance than algorithms without using TFA information (N-X-X) or algorithms using PLS based TFA prediction (P-X-X). The algorithms using APMI relevance score function (the right half of the plot) show slightly better performance than the algorithms using Pearson correlation relevance score function (the left half). And there are no significant differences due to the use of the CLR background correction (X-X-C or X-X-N).