9. Model building with ModelAngelo

As of release 5.0, RELION comes with a machine-learning approach for automated atomic model building called ModelAngelo. ModelAngelo is a graph neural network that combines information from the cryo-EM map with sequence information of the proteins that are present in the map to build an atomic model.

9.1. Download FASTA sequence

Because we know that the sample in this data set is beta-galactosidase from E.coli, we can provide the protein sequence as a FASTA file. Select Fetch sequence job from Fetch category and fill in following parameters:

UNIPROT accession number: P00722

Press RUN. We don’t need to inspect the RESULTS as it will only contain a .fasta text file containing the sequence string.

9.2. Model building

Now select the ModelAngelo job from Atomic Model Build category and fill in following parameters:

B-factor sharpened map:: PostProcess/jobXXX/postprocess_masked.mrc

FASTA sequence for proteins:: Fetch/jobXXX/P00722.fasta

Which GPU to use:: <leave blank>

Press RUN. The job should take about 10 minutes to run. Once it’s finished we can inspect the RESULTS tab. Adjust the Iso Value and inspect the model. The model should fit nicely in the density.

Please note that this model isn’t finished. This should be considered as the first step towards model building - obtaining an initial model. It is crucial to complete a number of Model Refinement and Model Validation steps. While these are out of scope for this tutorial, a tutorial for Model Building can be found HERE.