(If you manage a computational cluster, please skip to the next section)

Step 1: Install Conda

To install ModelAngelo, you need Anaconda. We recommend installing miniconda3, as it is lighter. You can find that here: miniconda

Once you have a version of Anaconda installed, check that it actually runs with:

conda info

Step 2: Clone this repo

Now, you can install ModelAngelo. First, you need to clone this Github repository with

git clone https://github.com/3dem/model-angelo.git

Step 3: Run install script

After, all you need to do is go into the model-angelo directory and run the install script:

cd model-angelo
source install_script.sh

You will now have a conda environment called model_angelo that is able to run the program. You need to activate this conda environment with conda activate model_angelo. Now, you can run model_angelo build -h to see if the installation worked!

If you manage a computational cluster with many users and would like to install ModelAngelo once to be used everywhere, you should complete the above steps 1 and 2 for a public account.

Next, you should designate a folder to save the shared weights of the model such that it can be readable and executable by all users of your cluster. Let's say that path is /public/model_angelo_weights.

Now, you run the following:

export TORCH_HOME=/public/model_angelo_weights
cd model-angelo
source install_script.sh --download-weights

Once the script is finished running, make sure that where it installed the weights is in the directory you set.

Finally, you can make the following bash script available for all users to run:

#!/bin/bash
source `which activate` model_angelo
model_angelo "$@"

1. Binary activate not found It appears that miniconda's activate binary is not added to PATH by default. You can either fix this by appending it yourself, like so:

export PATH="$PATH:/path/to/miniconda3/bin"

or running conda init and restarting your shell.

This is the recommended use case, when you have access to a medium-high resolution cryo-EM map (resolutions exceeding 4 Å) as well as a FASTA files with all of your protein, RNA, and DNA sequences.

Let's say the map's name is map.mrc and the (protein) sequence file is prot.fasta. To build your model in a directory named output, you run:

model_angelo build -v map.mrc -pf prot.fasta -o output

If you would like to build nucleotides as well, you need to provide the RNA and DNA portions of your sequences in different files like so

model_angelo build -v map.mrc -pf prot.fasta -df dna.fasta -rf rna.fasta -o output

If you only have RNA or DNA, you can drop the other input.

If the output of the program halts before the completion of GNN model refinement, round 3 / 3, there was a bug that you can see in output/model_angelo.log. Otherwise, you can find your model in output/output.cif. The name of the mmCIF file is based on the output folder name, so if you specify, for example, -o testing/test/model_building, the model will be in testing/test/model_building/model_building.cif.

If you have a sample where you do not know all of the protein sequences that occur in the map, you can run model_angelo build_no_seq instead. This version of the program uses a network that was not trained with input sequences, nor does it do post-processing on the built map.

Instead, in addition to a built model, it provides you with HMM profile files that you can use to search a database such as UniRef with HHblits.

You run this command:

model_angelo build_no_seq -v map.mrc -o output

The model will be in output/output.cif as before.

Now there are also HMM profiles for each chain in HMMER3 format here: output/hmm_profiles.

To do a sequence search, you should first download a fasta file that includes your proteins of interest. Often, this will be the proteome of the organism in question. Here are some example proteomes in a FASTA format:

After downloading this file, you will need to uncompress it with gunzip. Then, you can run

model_angelo hmm_search --i output --f PATH_TO_DB --o hmm_output

You will have your results as a series of HMM output files with the extension .hhr, for example: hmm_output/A.hhr. These are named by chain according to the model built by ModelAngelo in output/output.cif.

For further convenience, especially for large maps, we also now have two csv files named output/all_hits.csv and output/best_hits.csv. The first will have all the hits for all the chains sorted by the E-value. The second has all the unique hits, with the information of the best hit.