Metric Ion Classification Tutorial

Installation

To install MIC:

git clone https://github.com/keiserlab/metric-ion-classification.git 
cd metric-ion-classification
conda env create --name mic-env -f mic-env.yml
conda activate mic-env
pip install .

Installation may take some time, especially regarding conda dependency solving. We strongly recommend using an updated conda with conda-libmamba-solver, in which case installation should take a few minutes.

If you encounter issues intalling LUNA, see https://github.com/keiserlab/LUNA installation instructions.

Requirements

The dependencies for MIC are listed in mic-env.yml, reproduced here:

  - rdkit
  - seaborn
  - openbabel
  - pip
  - pymol-open-source
  - xopen
  - colorlog
  - networkx
  - scipy
  - pytorch
  - cpuonly
  - numpy=1.21.6
  - biopython==1.72
  - mmh3==2.5.1
  - pdbecif
  - joblib
  - tqdm
  - scikit-learn
  - simplejson

System requirements:

• MIC has been successfully installed and tested on Ubuntu 18.04, 20.04, and 22.04, Mac OS Monterey through Sonoma, and on a Windows machine running Ubuntu through WSL.

Running MIC

Command line

To run:

mic_predict -o <outfile> -e <path/to/entry_file>[optional] </path/to/pdb_file>

Inputs

• pdb_file / pdb_dir : path to either 1) a single PDB file in the case of running one 1 file or 2) a directory containing multuple PDB files, used with --entries flag

Options:

• -e/--entries: Run individual examples from a single file or multiple PDB files. See below.
• -n/--num_processes: Number of concurrent processes to use for FP generation - default 8
• -o/--outfile: CSV created to store generated fingerprints or MIC predictions. If not provided, output is displayed in the terminal.
• -p/--preds_only: Only output predictions, not full probabilities from SVC.
• -ext/--extended_set: Generate predictions for the extended set of labels (K, Mn,Iod,Fe,Br)
• -cpu/--cpu_only: Ignore available GPUs and generate all predictions on CPU.
• -co/--symexp_cutoff: Radius in angstroms around densities kept following symmetry expansion. If running a single PDB file, it is recommended to use 7 for X-ray crystallography or -1 for Cryo-EM structures. Only use other values if you are planning on training a new model - default 7

The following options are NOT recommended to change from their defaults, unless intending to train a new model. Do not change these if running predictions alone.

• -t/--fp_type: Type of fingerprint to generate and use, options are ['prune-eifp']
• -l/--length: length of the fingerprint vector to be generated - default 4096
• -sn/--shell_number: number of shells to use during fingerprint generation - default 18
• -sr/--shell_radius: shell radius to use during fingerprint generation - default 0.25
• -ub/--unblinded: flag to not zero the invariants of spherical densities
• -b/--bit: whether to generate count fingerprints or bit fingerprints - default is count

Entries

Entries are used to generate predictions for specific densities within a single PDB file or across multiple PDB files in one MIC run. To run with entries, the first argument to mic_predict should be a path to the directory containing all pdb files, while the entries argument should be a csv in the format 'pdb_id,chain,name,res_number,skip_symexpansion (must be on for Cryo-EM structures)':

4OKE,A,MG,202,
3A09,A,CA,601,
4L9P,B,ZN,601,
7aue,R,CA,501,X
7f53,R,CA,601,X
7f55,R,CA,601,X
7f54,R,CA,601,X
7f58,R,CA,602,X
7piu,R,CA,401,X
7piv,R,CA,401,X

Testing

To confirm correct installation:

pip install pytest
python -m pytest

The tests may take a few minutes to run.

ChimeraX Bundle

Actively under development, coming soon!

Citing MIC

If you use this software, please cite our preprint.