cg2all

Convert coarse-grained protein structure to all-atom model

Web server / Google Colab notebook

A demo web page is available for conversions of CG model to all-atom structure via Huggingface space.

A Google Colab notebook is available for tasks: - Task 1: Conversion of an all-atom structure to a CG model using convert_all2cg - Task 2: Conversion of a CG model to an all-atom structure using convert_cg2all - Task 3: Conversion of a CG simulation trajectory to an atomistic simulation trajectory using convert_cg2all

A Google Colab notebook is available for local optimization of a protein model structure against a cryo-EM density map using cryoemminimizer.py

Installation

These steps will install Python libraries including cg2all (this repository), a modified MDTraj, a modified SE3Transformer, and other dependent libraries. The installation steps also place executables convert_cg2all and convert_all2cg in your python binary directory.

This package is tested on Linux (CentOS) and MacOS (Apple Silicon, M1).

for CPU only

bash pip install git+http://github.com/huhlim/cg2all

for CUDA (GPU) usage

1. Install Miniconda
2. Create an environment with DGL library with CUDA support bash # This is an example with cudatoolkit=11.3. # Set a proper cudatoolkit version that is compatible with your CUDA driver and DGL library. # dgl>=1.1 occasionally raises some errors, so please use dgl<=1.0. conda create --name cg2all pip cudatoolkit=11.3 dgl=1.0 -c dglteam/label/cu113
3. Activate the environment bash conda activate cg2all
4. Install this package bash pip install git+http://github.com/huhlim/cg2all

for cryoemminimizer usage

You need additional python package, mrcfile to deal with cryo-EM density map. bash pip install mrcfile

Usages

convert_cg2all

convert a coarse-grained protein structure to all-atom model ```bash usage: convertcg2all [-h] -p INPDBFN [-d INDCDFN] -o OUTFN [-opdb OUTPDBFN] [--cg {supportedcgmodels}] [--chain-break-cutoff CHAINBREAKCUTOFF] [-a] [--fix] [--ckpt CKPTFN] [--time TIME_JSON] [--device DEVICE] [--batch BATCHSIZE] [--proc NPROC]

options: -h, --help show this help message and exit -p INPDBFN, --pdb INPDBFN -d INDCDFN, --dcd INDCDFN -o OUTFN, --out OUTFN, --output OUTFN -opdb OUTPDBFN --cg {supportedcgmodels} --chain-break-cutoff CHAINBREAKCUTOFF -a, --all, --isall --fix, --fixatom --standard-name --ckpt CKPTFN --time TIMEJSON --device DEVICE --batch BATCHSIZE --proc NPROC ```

arguments

• -p/--pdb: Input PDB file (mandatory).
• -d/--dcd: Input DCD file (optional). If a DCD file is given, the input PDB file will be used to define its topology.
• -o/--out/--output: Output PDB or DCD file (mandatory). If a DCD file is given, it will be a DCD file. Otherwise, a PDB file will be created.
• -opdb: If a DCD file is given, it will write the last snapshot as a PDB file. (optional)
• --cg: Coarse-grained representation to use (optional, default=CalphaBasedModel).
  • CalphaBasedModel: CA-trace (atom names should be "CA")
  • ResidueBasedModel: Residue center-of-mass (atom names should be "CA")
  • SidechainModel: Sidechain center-of-mass (atom names should be "SC")
  • CalphaCMModel: CA-trace + Residue center-of-mass (atom names should be "CA" and "CM")
  • CalphaSCModel: CA-trace + Sidechain center-of-mass (atom names should be "CA" and "SC")
  • BackboneModel: Model only with backbone atoms (N, CA, C)
  • MainchainModel: Model only with mainchain atoms (N, CA, C, O)
  • Martini: Martini model
  • Martini3: Martini3 model
  • PRIMO: PRIMO model
• --chain-break-cutoff: The CA-CA distance cutoff that determines chain breaks. (default=10 Angstroms)
• --fix/--fix_atom: preserve coordinates in the input CG model. For example, CA coordinates in a CA-trace model will be kept in its cg2all output model.
• --standard-name: output atom names follow the IUPAC nomenclature. (default=False; output atom names will use CHARMM atom names)
• --ckpt: Input PyTorch ckpt file (optional). If a ckpt file is given, it will override "--cg" option.
• --time: Output JSON file for recording timing. (optional)
• --device: Specify a device to run the model. (optional) You can choose "cpu" or "cuda", or the script will detect one automatically.
  "cpu" is usually faster than "cuda" unless the input/output system is really big or you provided a DCD file with many frames because it takes a lot for loading a model ckpt file on a GPU.
• --batch: the number of frames to be dealt at a time. (optional, default=1)
• --proc: Specify the number of threads for loading input data. It is only used for dealing with a DCD file. (optional, default=OMPNUMTHREADS or 1)

examples

Conversion of a PDB file bash convert_cg2all -p tests/1ab1_A.calpha.pdb -o tests/1ab1_A.calpha.all.pdb --cg CalphaBasedModel Conversion of a DCD trajectory file bash convert_cg2all -p tests/1jni.calpha.pdb -d tests/1jni.calpha.dcd -o tests/1jni.calpha.all.dcd --cg CalphaBasedModel Conversion of a PDB file using a ckpt file bash convert_cg2all -p tests/1ab1_A.calpha.pdb -o tests/1ab1_A.calpha.all.pdb --ckpt CalphaBasedModel-104.ckpt

convert_all2cg

convert an all-atom protein structure to coarse-grained model ```bash usage: convertall2cg [-h] -p INPDBFN [-d INDCDFN] -o OUTFN [--cg {supportedcgmodels}]

options: -h, --help show this help message and exit -p INPDBFN, --pdb INPDBFN -d INDCDFN, --dcd INDCDFN -o OUTFN, --out OUTFN, --output OUT_FN --cg ```

arguments

• -p/--pdb: Input PDB file (mandatory).
• -d/--dcd: Input DCD file (optional). If a DCD file is given, the input PDB file will be used to define its topology.
• -o/--out/--output: Output PDB or DCD file (mandatory). If a DCD file is given, it will be a DCD file. Otherwise, a PDB file will be created.
• --cg: Coarse-grained representation to use (optional, default=CalphaBasedModel).
  • CalphaBasedModel: CA-trace (atom names should be "CA")
  • ResidueBasedModel: Residue center-of-mass (atom names should be "CA")
  • SidechainModel: Sidechain center-of-mass (atom names should be "SC")
  • CalphaCMModel: CA-trace + Residue center-of-mass (atom names should be "CA" and "CM")
  • CalphaSCModel: CA-trace + Sidechain center-of-mass (atom names should be "CA" and "SC")
  • BackboneModel: Model only with backbone atoms (N, CA, C)
  • MainchainModel: Model only with mainchain atoms (N, CA, C, O)
  • Martini: Martini model
  • Martini3: Martini3 model
  • PRIMO: PRIMO model

an example

bash convert_all2cg -p tests/1ab1_A.pdb -o tests/1ab1_A.calpha.pdb --cg CalphaBasedModel

script/cryoemminimizer.py

Local optimization of protein model structure against given electron density map. This script is a proof-of-concept that utilizes cg2all network to optimize at CA-level resolution with objective functions in both atomistic and CA-level resolutions. It is highly recommended to use cuda environment. ```bash usage: cryoemminimizer [-h] -p INPDBFN -m INMAPFN -o OUT_DIR [-a] [-n N_STEP] [--freq OUTPUTFREQ] [--chain-break-cutoff CHAINBREAKCUTOFF] [--restraint RESTRAINT] [--cg {CalphaBasedModel,CA,ca,ResidueBasedModel,RES,res}] [--standard-name] [--uniformrestraint] [--nonuniformrestraint] [--segment SEGMENTS]

options: -h, --help show this help message and exit -p INPDBFN, --pdb INPDBFN -m INMAPFN, --map INMAPFN -o OUTDIR, --out OUTDIR, --output OUTDIR -a, --all, --isall -n NSTEP, --step NSTEP --freq OUTPUTFREQ, --outputfreq OUTPUTFREQ --chain-break-cutoff CHAINBREAKCUTOFF --restraint RESTRAINT --cg {CalphaBasedModel,CA,ca,ResidueBasedModel,RES,res} --standard-name --uniformrestraint --nonuniformrestraint --segment SEGMENTS ```

arguments

• -p/--pdb: Input PDB file (mandatory).
• -m/--map: Input electron density map file in the MRC or CCP4 format (mandatory).
• -o/--out/--output: Output directory to save optimized structures (mandatory).
• -a/--all/--is_all: Whether the input PDB file is atomistic structure or not. (optional, default=False)
• -n/--step: The number of minimization steps. (optional, default=1000)
• --freq/--output_freq: The interval between saving intermediate outputs. (optional, default=100)
• --chain-break-cutoff: The CA-CA distance cutoff that determines chain breaks. (default=10 Angstroms)
• --restraint: The weight of distance restraints. (optional, default=100.0)
• --cg: Coarse-grained representation to use (default=ResidueBasedModel)
• --standard-name: output atom names follow the IUPAC nomenclature. (default=False; output atom names will use CHARMM atom names)
• --uniformrestraint/--nonuniformrestraint: Whether to use uniform restraints. (default=True) If it is set to False, the restraint weights will be dependent on the pLDDT values recorded in the PDB file's B-factor columns.
• --segment: The segmentation method for applying rigid-body operations. (default=None)
  • None: Input structure is not segmented, so the same rigid-body operations are applied to the whole structure.
  • chain: Input structure is segmented based on chain IDs. Rigid-body operations are independently applied to each chain.
  • segment: Similar to "chain" option, but the structure is segmented based on peptide bond connectivities.
  • 0-99,100-199: Explicit segmentation based on the 0-index based residue numbers.

an example

bash ./cg2all/script/cryo_em_minimizer.py -p tests/3isr.af2.pdb -m tests/3isr_5.mrc -o 3isr_5+3isr.af2 --all

Datasets

The training/validation/test sets are available at zenodo.

Reference

Lim Heo & Michael Feig, "One particle per residue is sufficient to describe all-atom protein structures", bioRxiv (2023). Link