FIGURE LEGENDS

Figure 1. Prediction pipeline schematic and example targets for manual interventions. (A) A schematic representation of the prediction pipeline. The green blocks represent steps used in the default AF2 pipeline. Blocks with other colors represent modifications to the pipeline. (B) An example of targets for which MSA arrangements were performed. The y-axis represents MSA depth, and the x-axis represents sequence position. (C) The MSA shown in B after the manual arrangement. (D) An example of targets exhibiting a globular structure after the refinement. (E) The refined result of the structure shown in D. MSA, multiple sequence alignment.
Figure 2. Prediction conditions for T1173. “Semi-auto” represents MSA results from the standard pipeline outlined in the Materials and Methods section. “ExMSA” refers to MSA generated with an added 196 aa at the N-terminus. “ExMSA (filtered)” denotes the ExMSA after the MSA arrangement process detailed in the main text. (A) MSA-depth pre- and post-manual interventions. (B) MSA depth with a focus on y-axis values below 2000. (C) Predicted structure examples. Note that the TM score was unavailable during the prediction season. “ColabFold” represents the model downloaded from the ColabFold website (https://wwwuser.gwdg.de/~mmirdit/casp15/). MSA, multiple sequence alignment; aa, amino acid.
Figure 3. Accuracy differences between structures predicted using MSAs generated by the PEZYFoldings pipeline (PEZY-MSA) and those provided by the NBIS-AF2-standard and NBIS-AF2-multimer teams (NBIS-MSA). (A) TM-scores of MODEL 1 (the model with the highest confidence) for each single-domain target. (B) TM-scores of the best structures for each single-domain target. (C) ΔTM-score (TM-score of models using PEZY-MSA minus TM-score of models using NBIS-MSA) of MODEL 1 as a function of Nseq-NBIS-MSA (number of sequences in NBIS-MSA). (D) ΔTM-score of the best models as a function of Nseq-NBIS-MSA. (E) CASP15 official Z-M1-GDT (Z-scores of MODEL 1 based on GDT-TS) for the single-domain category submitted by PEZYFoldings as a function of Nseq-NBIS-MSA. The results of the 52 targets used in this analysis were plotted. The results of T1145-D1 and T1145-D2 were not plotted in E because I did not submit MODEL 1 for this target.
Figure 4. Comparison of TM-scores for predicted structures before and after refinement. (A) TM-scores for all submitted structures for single-domain targets processed by the refinement model. (B) MODEL 1 structures’ TM-scores for single-domain targets. (C) Highest TM-score among five submitted structures for single-domain targets. (D) TM-scores for all submitted structures for multimer targets processed by the refinement model. (E) MODEL 1 structures’ TM-scores for multimer targets (F) Highest TM-score among five submitted structures for multimer targets.