CG2AT2 — Lite Version

Citation: Vickery ON, Stansfeld PJ. CG2AT2: an Enhanced Fragment-Based Approach for Serial Multi-scale Molecular Dynamics Simulations. J Chem Theory Comput. 2021;17(10):6472–6482. doi:10.1021/acs.jctc.1c00295

---

Table of Contents

1. Overview
2. Installation
3. Requirements
4. Quick Start
5. Flags Reference
6. How It Works
7. Input
8. Output Structure
9. Output Conversion Modes
10. Detailed Flag Guide
11. Database Structure
12. Adding Fragments to the Database

---

Overview

CG2AT2 converts Coarse-Grained (CG) molecular systems (e.g. Martini) back to atomistic resolution using a fragment-based fitting approach. It is designed to require minimal input from the user — in most cases only the CG coordinate file and the original atomistic protein structure are needed, and CG2AT2 produces all files required to run further atomistic simulations.

---

Installation

CG2AT2 requires no compilation. Download it directly from GitHub and run it immediately.

Recommended: add CG2AT2 to your PATH for convenience by appending the following to your ~/.profile or ~/.bashrc:

export PATH="/path/to/CG2AT2:$PATH"

---

Requirements

Category	Requirement
Python	v3.0 or higher
GROMACS	v5 or higher
NumPy	any recent version
SciPy	any recent version

All other dependencies (argparse, copy, glob, math, multiprocessing, os, pathlib, re, shutil, subprocess, sys, time, etc.) are included in the Python standard library.

Note: distutils is no longer required as of this release.

---

Quick Start

Minimal run (CG structure only):

python cg2at.py -c cg_input.gro

Recommended run (with original atomistic structure for improved quality):

python cg2at.py -c cg_input.gro -a atomistic_input.gro

Fully automated run (no interactive prompts):

python cg2at.py -c cg_input.gro -a atomistic_input.gro \
    -w tip3p -fg martini_2-2_charmm36 -ff charmm36-jul2022

---

Flags Reference

Required

Flag	Type	Description
-c	pdb/gro/tpr	Coarse-grained input coordinates

Optional

Flag	Type	Default	Description
-a	pdb/gro/tpr	—	Atomistic input coordinates
-d	0:2 1:3	—	Duplicate atomistic chains (chain:copies)
-loc	str	CG2AT_<timestamp>	Output folder name
-o	all/align/de_novo/none	all	Output conversion mode
-group	0,1 2,3 / all / chain	—	Rigid-body fitting groups
-ncpus	int	8 (or max)	Number of CPU cores
-mod	flag	off	Treat fragments individually (disable grouping)
-sf	float	0.9	Fragment scale factor before fitting
-cys	float	7.0	Disulphide bond search cutoff (Å)
-ter	flag	off	Interactively choose terminal species
-nt	flag	off	Neutral N-terminus on all chains
-ct	flag	off	Neutral C-terminus on all chains
-vs	flag	off	Apply virtual sites to protein
-swap	str	—	Swap residues or beads during conversion
-box	float float float	—	New box dimensions in Å (0 = keep original)
-w	str	—	Water model (e.g. tip3p, spce)
-ff	str	—	Force field (e.g. charmm36-jul2022)
-fg	str	—	Fragment library (e.g. martini_2-2_charmm36)
-gmx	str	gmx	GROMACS executable name
-messy	flag	off	Keep all temporary files
-silent	flag	off	Auto-accept disulphide bond suggestions
-disre	flag	off	Disable backbone distance restraint matrix
-ov	float	0.3	Minimum allowed atom overlap (Å)
-posre	str	—	Generate position restraint file for a residue
-compare	str	—	Compare an .itp file against the database
-info	flag	off	Print available force fields and fragments
-version	flag	off	Print CG2AT2 version
-v	flag (stackable)	off	Increase output verbosity (up to -vvv)
-h	flag	—	Show help and exit

---

How It Works

CG2AT2 uses a fragment-based fitting workflow. Each fragment is placed and rotated independently with no knowledge of neighbouring beads:

1. Centre each fragment at the centre of mass (COM) of its heavy atoms.
2. Rotate the fragment to minimise the distance between atoms that connect to adjacent beads.
3. Minimise each residue individually.
4. Merge all residues and minimise the complete system.
5. Check for threaded lipids (bonds > 0.2 nm) and correct them.
6. Run NVT equilibration (where applicable).
7. Morph the protein to the user-supplied atomistic structure via steered MD (where applicable).

Threaded Lipid Correction

Lipid tails can accidentally thread through aromatic residues during CG-to-AT conversion. CG2AT2 detects this by checking all bond lengths in the minimised system. Any bond exceeding 0.2 nm is treated as a threading event — the offending atoms are corrected and the system is re-minimised.

---

Input

CG2AT2 supports two rebuilding strategies:

• De novo: Fragments are fitted to the CG bead positions from scratch.
• Flexible fitting: A user-supplied atomistic structure is sequence-aligned to the CG system and then morphed into place via steered MD. This produces the highest quality output and should be used whenever the original atomistic structure is available.

If only a partial atomistic structure is supplied (e.g. the main protein but not a flexible linker), CG2AT2 uses the de novo method to build in the missing regions and seamlessly combines the two.

Multiple atomistic input files are accepted:

python cg2at.py -c cg_input.gro -a chain_A.pdb chain_B.pdb

---

Output Structure

CG2AT_<timestamp>/
├── INPUT/
│   ├── CG_INPUT.pdb          # CG structure converted to PDB
│   ├── AT_INPUT_X.pdb        # Supplied atomistic structure(s) as PDB
│   └── script_inputs.dat     # All flags used, saved for reproducibility
│
├── <RESIDUE_TYPE>/           # One folder per residue type (e.g. PROTEIN, POPE)
│   ├── *.pdb                 # Individually converted residues
│   ├── *.top / *.itp         # Residue topologies
│   ├── *.mdp                 # Minimisation run files
│   ├── gromacs_outputs       # Full GROMACS log
│   └── MIN/                  # Minimised residue PDBs
│
├── MERGED/
│   ├── merged_cg2at_*.pdb    # All residue types merged into one PDB
│   ├── *.top / *.itp         # Combined topology
│   ├── MIN/                  # Merged minimisation files
│   ├── NVT/                  # NVT equilibration files
│   └── STEER/                # Steered MD alignment files
│
└── FINAL/
    ├── <forcefield>.ff/      # Force field directory
    ├── *.top / *.itp         # Final topology files
    ├── final_cg2at_de_novo.pdb     # De novo output
    ├── final_cg2at_aligned.pdb     # Aligned output (if applicable)
    ├── script_timings.dat    # Per-stage runtime breakdown
    └── RMSD data             # RMSD between CG and atomistic output

---

Output Conversion Modes

Select the output mode with -o. The default is all.

Mode	NVT	Steered MD	Output file(s)
none	✗	✗	final_cg2at_de_novo.pdb
de_novo	✓ (5 ps)	✗	final_cg2at_de_novo.pdb
align	✗	✓	final_cg2at_aligned.pdb
all	✓ (5 ps)	✓	both of the above

All modes start with fragment fitting, minimisation, and threaded-lipid correction.

---

Detailed Flag Guide

Disulphide Bonds

CG2AT2 searches for disulphide bonds in both the user atomistic structure (S–S < 2.1 Å) and the CG representation (SC1–SC1 < 7 Å, more than 4 residues apart). When a bond is found only in the CG structure it will prompt for confirmation.

-silent          # Auto-accept all disulphide suggestions without prompting
-cys 10.0        # Expand search radius to 10 Å for stretched martini bonds

If you see a topology/atom-count mismatch that is off by exactly 2, it is almost always an undetected disulphide bond — try increasing -cys.

---

Residue and Bead Swapping

The -swap flag allows simple mutations or residue substitutions during conversion, without editing the input file. The general syntax is:

-swap FROM_residue,bead:TO_residue,bead:resid_range

Use case	Example
Swap all ASP to ASN	-swap ASP:ASN
Swap ASP to ASN in resid range 0–10 and 30–40	-swap ASP:ASN:0-10,30-40
Swap residues with different bead names	-swap POPC,NC3:POPG,GL0
Swap beads within the same residue	-swap POPG,D2B:POPG,C2B
Skip a bead	-swap GLU,SC2:ASP,skip
Skip an entire residue	-swap POPG:skip
Multiple swaps at once	-swap POPE,NH3:POPG,GL0 POPG,GL0:POPE,NH3
Skip specific NA+ ions by resid	-swap NA+:skip:4000-4100

Correcting residue name truncation: PDB and GRO formats truncate residue names to four characters, which causes ambiguity with closely related residues such as phosphoinositides. CG2AT2 internally supports longer names. For example, the three PI headgroup variants are stored as POPI1_3, POPI1_4, and POPI1_5, and a four-character POPI bead in the input can be redirected:

-swap POPI:POPI1_4

---

Protein Chain Fitting

For multimeric proteins, CG2AT2 offers several rigid-body fitting strategies:

# Default: fit each atomistic chain to its CG counterpart independently
python cg2at.py -c cg.gro -a protein.pdb

# Fit by CG chain (one group per CG chain)
python cg2at.py -c cg.gro -a protein.pdb -group chain

# Fit the entire atomistic structure as a single rigid body
python cg2at.py -c cg.gro -a protein.pdb -group all

# Fit specific chains as groups (chains 0+2 together, chains 1+3 together)
python cg2at.py -c cg.gro -a protein.pdb -group 0,2 1,3

---

Chain Duplication

For homomeric complexes only one chain needs to be supplied; the -d flag copies it:

# Use 3 copies of chain 0 and 2 copies of chain 1
python cg2at.py -c cg.gro -a monomer.pdb -d 0:3 1:2

---

Terminal Residues

CG2AT2 uses charged termini by default (most force fields require this). Override with:

-nt       # Neutral N-terminus on all chains
-ct       # Neutral C-terminus on all chains
-ter      # Interactive selection per chain

---

Fragment Scale Factor

Fragments are shrunk to 90% of their original size before fitting, reducing atom clashes. Decrease further if minimisation fails:

-sf 0.8   # Shrink to 80%

---

Fragment Grouping

By default, adjacent beads within a residue are fitted as a group, improving geometry (especially for sugars). To disable:

-mod      # Treat every bead fragment independently

---

PBC Box Resizing

Box resizing is currently supported for cubic boxes only. Dimensions are in Ångströms; use 0 to preserve the original value on a given axis:

-box 100 100 100    # Set all axes to 100 Å
-box 0 0 100        # Shrink only the Z-axis to 100 Å

---

Distance Restraint Matrix

When an atomistic structure is supplied, CG2AT2 generates a backbone hydrogen-bond distance restraint matrix that is applied during NVT. This refines backbone atom orientations with minimal effect on overall structure. Disable with:

-disre

---

Atom Overlap Checker

Atoms closer than 0.3 Å will cause minimisation to fail. Increase the cutoff if needed:

-ov 0.5   # Allow up to 0.5 Å overlap

---

Resuming a Failed Run

CG2AT2 skips any step for which output files already exist, so a failed run can be resumed after fixing the error simply by re-running the same command with the same -loc directory:

-loc CG2AT   # Use a fixed directory name instead of a timestamp

---

Miscellaneous

-gmx gmx_mod     # Use a specific GROMACS binary
-messy           # Keep all temporary files
-vs              # Apply virtual sites to protein hydrogen atoms
-ncpus 4         # Limit to 4 CPU cores (default: 8 or system max)
-v / -vv / -vvv  # Increase verbosity
-info            # List available force fields and fragment libraries
-info -fg martini_2-2_charmm36   # List residues in a specific library
-posre POPE      # Generate a position restraint file for POPE
-compare martini_v3.itp   # Compare an itp against the database

---

Database Structure

database/
├── forcefields/
│   └── charmm36.ff/          # Standard GROMACS force field directory
│
└── fragments/
    └── <forcefield_type>/    # e.g. martini_2-2_charmm36
        ├── protein/
        │   └── <AA>/         # e.g. ASP, PHE
        │       ├── <AA>.pdb  # Fragment (bead sections)
        │       └── <AA>.top  # Topology (optional)
        ├── protein_modified/
        │   └── <residue>/    # e.g. CYSD
        ├── non_protein/
        │   └── <residue>/    # e.g. POPC
        │       ├── <RES>.pdb
        │       ├── <RES>.itp
        │       └── <RES>.top (optional)
        ├── solvent/
        │   └── <bead>/       # e.g. W
        │       └── TIP3P.pdb, SPCE.pdb, ...
        ├── ions/
        │   └── <ion>/        # e.g. NA+
        └── other/
            └── <residue>/    # e.g. DA (DNA)

Tip: Prefix any file or folder name with _ to prevent CG2AT2 from reading it.

---

Adding Fragments to the Database

Fragment PDB Format

Each fragment PDB uses [ bead_name ] section headers followed by standard ATOM records:

[ BB ]
ATOM      1  N   PHE     1      42.030  16.760  10.920  2.00  0.00           N
ATOM      2  CA  PHE     1      42.770  17.920  11.410  3.00  1.00           C
ATOM     10  C   PHE     1      44.240  17.600  11.550  2.00  0.00           C
ATOM     11  O   PHE     1      44.640  16.530  12.080  6.00  0.00           O
[ SC1 ]
ATOM      3  CB  PHE     1      42.220  18.360  12.800  1.00  1.00           C
...

• Protein amino acids should not include adjustable hydrogens — these are added automatically by pdb2gmx.
• Modified protein and non-protein fragments should include all hydrogens, as they are not processed by pdb2gmx.

Optional Topology File

A .top file of the same name can provide connectivity, grouping, chirality, and hydration information:

[ CONNECT ]
# bead_1  atom_1  bead_2  direction
   BB       N      BB        -1
   BB       C      BB         1

[ GROUPS ]
SC1 SC2 SC3

[ CHIRAL ]
# central  move  atom1  atom2  atom3
   CA       HA    CB     N      C

[ HYDRATION ]
W

Solvent Fragments

Each solvent model is a separate PDB file inside the bead folder (e.g. solvent/W/TIP3P.pdb). A single CG water bead can expand to multiple atomistic molecules:

[ TIP3P ]
ATOM      1  OW  SOL     1      20.910  21.130  75.300  1.00  0.00
ATOM      2  HW1 SOL     1      20.580  21.660  76.020  1.00  0.00
ATOM      3  HW2 SOL     1      21.640  21.640  74.940  1.00  0.00
ATOM      4  OW  SOL     2      21.000  22.960  77.110  1.00  0.00
...

Ion Fragments

Each ion is stored as a single-atom fragment. The [ HYDRATION ] topology entry controls which solvent bead is overlaid to model the first hydration shell:

[ NA+ ]
ATOM      1  NA   NA     1      21.863  22.075  76.118  1.00  0.00

[ HYDRATION ]
W

Generating Position Restraint Files

CG2AT2 can generate a position restraint .itp for any non-protein residue in the database:

python cg2at.py -posre POPE -fg martini_2-2_charmm36

This creates POPE_posre.itp and adds the corresponding #ifdef NP block to the residue's .itp file.