LiveBench-6: Large-Scale Automated Evaluation of Protein Structure Prediction Servers

The aim of the LiveBench experiment is to provide a continuous evaluation of structure prediction servers in order to inform potential users about the current state-of-the-art structure prediction tools and in order to help the developers to analyze and improve the services. This round of the experiment was conducted in parallel to the blind CAFASP-3 evaluation experiment. The data collected almost simultaneously enables the comparison of servers on two different benchmark sets. The number of servers has doubled from the last evaluated LiveBench-4 experiment completed in April 2002, just before the beginning of CAFASP-3. This can be partially attributed to the rapid development in the area of meta-predictors (consensus servers). The current results confirm the high sensitivity and specificity of the meta-predictors. Nevertheless, the comparison between the autonomous (not meta) servers participating in the last CAF-ASP-2 and LiveBench-2 experiment and the current set of autonomous servers demonstrates that progress has been made also in sequence structure fitting functions. In addition to the growing number of participants, the current experiment marks the introduction of new evaluation procedures, which are aimed to correlate better with functional characteristics of models.