Figure 5

GTRNetwork algorithm combinations on input initial network of 70% RegulonDB 7.0 data. 30% of links randomly deleted. Five runs were made for each recall level. The trend lines of data points are fitted by polynomial functions. Under this condition the combination E-A-C (EM-based TFA prediction, APMI relevance score with CLR background correction) and E-A-N (EM-based TFA prediction, APMI relevance score without CLR background correction) give the best performances. All the TFA based algorithms except the SIMPLS based TFA prediction show significantly better performance than the algorithms not using TFA information. At the low recall levels, the regression based TFA prediction algorithms (P-C-C and P-A-C) have better performance than the algorithms not using TFA information.