Structure accuracy calculation (SAC)

The SAC assesses the accuracy of atomic models:

1. Each atomic model in the predicted ensemble is minimum root mean square deviation (min-RMSD) aligned to every atomic model in the ground-truth ensemble.
2. The alignment with the min-RMSD is taken as the optimal alignment for each atomic model in the predicted ensemble.
3. The SAC is the average min-RMSD for all the optimally-aligned models in the predicted ensemble.

$$ SAC=\frac{1}{N^{pred}{models}} \sum^{N^{pred}{models}}{i=1} \min{j}(RMSD(M_i^{pred}, M_{j}^{GT})) $$

where:

• $i\in{N^{pred}_{models}}$ is the number of models in the predicted ensemble
• $j\in{N^{GT}_{models}}$ is the number of models in the ground-truth ensemble
• $M_i^{pred}$ is the $i$’th atomic model in the predicted ensemble
• $M_{j}^{GT}$ is the $j$’th atomic model in the ground-truth ensemble

Local flexibility calculation (LFC)

The LFC assesses how similar local heterogeneity is to the ground truth:

1. The atomic models in the predicted ensemble are optimally aligned to the closed ground model via min-RMSD as in points 1+2 in the SAC section.
2. The root mean square fluctuation (RMSF) is then calculated for each residue for the optimally aligned ensemble and for the ground-truth ensemble.
3. For each residue, the RMSF of the predicted ensemble is subtracted from the RMSF of the ground truth ensemble.
4. The absolute value of the difference in RMSF for each residue is taken.
5. The LFC is the absolute value of the difference in RMSF averaged over all residues.

$$ LFC = \langle | RMSF_{GT}-RMSF_{pred} |\rangle $$

where |.| is the absolute value calculated for each residue, and the average is calculated over all residues.