Figure 2.

Correlation values for each function. Increasing values of α place increasing weight on the predicted-best set of models. An α value of 0 is equivalent to Kendall’s τ, treating all models equally. Values of 0, 3, 5, 15, and 30 place half of the weight on the top 50%, 23%, 14%, 5%, and 2.3% of models. The target correlation plot shows the average τ_α values over all targets, with correlation computed separately for each target. The global correlation plot shows the τ_α values computed from combining all predictions into a single set. Adding Undertaker cost function terms to the consensus median GDT_TS method improved the ranking of models within a target, particularly when concentrating on the top-scoring models. Median GDT_TS alone is a better predictor of raw GDT_TS value, especially for picking out the easy targets, but does not do as well at ranking models for a given target. Furthermore, the alignment based constraints do better globally than the Undertaker cost functions (which include the alignment based constraints) for ranking the most accurate models; thus, the quality of these models can be better judged by consistency with alignments alone. “Alignment Constraints” is MQAO, “Undertaker” is MQAU, “Undertaker+Consensus” is MQAC, and “Median GDT_TS” is the pure consensus term, which we did not submit to CASP8.