Fig. 2: Comparison of model performance on CAMEO and CASP15 targets.
From: An end-to-end framework for the prediction of protein structure and fitness from single sequence
a The boxplot compares the TM-scores of structures predicted by SPIRED, SPIRED+GDFold2, OmegaFold, and ESMFold with Cycle = 1 on the CAMEO set, which is composed of 680 samples of proteins. Here, the green triangle indicates the mean value with the corresponding number marked aside, the center line represents the median, the box limits correspond to the upper and lower quartiles and the whiskers extend to 1.5 inter-quartile range. A more detailed comparison is shown in the scatter plot, where each point represents a protein sample, the vertical axis represents the TM-score of SPIRED+GDFold2 predictions, and the horizontal axis represents the results of OmegaFold and ESMFold, respectively. b Similarly, the boxplot and scatter plot present the general and detailed comparisons between the evaluation results for these models with Cycle = 1 on the CASP15 set, which is composed of 45 samples of protein domains, whose sequences and ground truth structures are available from the CASP15 official website. The boxplot is constructed in the same way as that in subplot a. In both boxplots, the single-sequence-based predictors under evaluation are also compared with the standard AlphaFold2 (ColabFold67 version, Cycle = 1) taking the MSA or single sequence as input, respectively, as well as an ultrafast predictor EMBER3D68 (unrelaxed version). Source data are provided as a Source Data file.
