Refine structures to near-atomic resolution in minutes on a single GPU*Scale bar: 25 Å. Showing PDB1YAR docked into a cryoSPARC density map.*
High-speed refinement in cryoSPARC
The algorithms underlying cryoSPARC's refinement feature enable rapid refinement of cryo-EM structures. For example, the Thermoplasma acidophilum 20S proteasome (EMPIAR-10025) can be refined to 2.8 Å, equal to the previously published resolution, in only 32 minutes on a single consumer-grade NVIDIA GPU.
These results are achieved using the new optimized GPU kernels implemented in cryoSPARC v0.2.47, which provide a significant speed-up over previous versions, and over the results reported in a recent Nature Methods paper about cryoSPARC.
We first perform a single-class ab-initio reconstruction by navigating to the 'Experiments' tab and clicking 'New Experiment'. The available experiment types are shown in purple and we select 'Ab-Initio', leaving the number of classes set to the default value of 1:
On the Parameters page, we click on 'Show Advanced Params' relating to Ab-Initio:
Next, we scroll down to the Symmetry parameter and enter a value of 'D7'. In general, it is not recommended that you enforce symmetry during ab-initio reconstruction in cryoSPARC, because often symmetry of a structure is not known in advance of reconstruction. However, for this particular dataset, it is advantageous to enforce symmetry because there are insufficient top-views of the proteasome structure, i.e. a significant degree of orientation bias.
Click 'Launch' and 'Enqueue' to commence the experiment.
Refinement to 2.8 Å in 32 min
A projection image of the result structure from the ab-initio experiment can be viewed on the 'Experiments' page. The status indicator will change to green to indicate when the experiment is complete:
To start a refinement of the structure produced during the ab-initio run, simply click 'Structure' below the experiment card. This selects the single structure on which we want to perform the refinement:
Note: It is also possible to include the particles corresponding to a particular class if you want to include them in the refinement; however, this is not required for a single-class refinement since we want to use all particles in the dataset. In addition, each structure result can be selected multiple times, for instance if the goal is to initiate a multi-class reconstruction with several classes starting from a single reference structure.
Next, click 'New Experiment' to display the experiments that can be performed on the selection - this time, Refinement is available and we select that from the list:
As was the case during the ab-initio reconstruction, we enforce D7 symmetry by entering the parameter value. Other than symmetry, we do not need to change any of the default parameters:
Click 'Launch' and 'Enqueue' to commence the experiment and view/download results and plots in real-time. For example, we can directly view the three axis-aligned slices of the real-space structure, and three axis-aligned slices of the Fourier-space structure:
The dynamic refinement mask automatically generated by cryoSPARC is also shown, in three axis-aligned slices. The mask is generated by thresholding the density and expanding the resulting mask with a minimum distance and soft cosine edge:
The gold-standard FSC plot reveals a 2.8 Å result at the FSC=0.143 level: