From: Fotis Baltoumas (
Date: Thu Jan 26 2017 - 09:07:10 CST

Hello Giota,

The version of Martini in those files is 2.1 for the proteins and 2.0
for the lipids. Note that starting one of the things Martini 2.2
changed was the names of the protein beads (BB instead of BAS for the
backbone, SC1-SC4 for side chains instead of SID/SI1-4) so, if you plan
to use analysis tools intended for GROMACS results, make the appropriate

Furthermore, the lipids follow the original mapping scheme, not the
current one (established with the lipidome papers) which has some
changes (for example, palmitoyl chains are now mapped using four CG
beads instead of the original 5).

Finally, the lipid parameters only have phospholipids and no
cholesterol, glycolipids, cardiolipins etc are included (although with a
bit of work you may transfer them from GROMACS yourself).

As for your second question, results from Martini simulations in NAMD
and GROMACS are generally the same. The paper of the CHARMM-GUI PACE CG
Builder compared non-bonded energies for NAMD and GROMACS Martini and
found virtually no changes among them. As for proteins, I haven't seen
any publications comparing the two schemes, but we have run some
comparisons in our lab and found that for the cases we studied, both
NAMD and GROMACS gave similar results (and replicated results from
all-atom runs too).

However, there are some things you should look out for:

1. In GROMACS Martini, temperature coupling is done through v-rescale,
while pressure is controlled either through the Berendsen bath or
through Parinello-Rahman. On the other hand, NAMD Martini simulations
in the tutorial use the Langevin Dynamics/Langevin piston combination
for NPT. Although this probably has minor effects on protein dynamics,
it can affect your global system properties, the simulation speed (NAMD
Martini is, generally, slower than GROMACS Martini) and possibly the
dynamics of a lipid bilayer (e.g. curvature). Also, the Langevin piston
often tends to crash Martini simulations, complaining that the periodic
box has become too small for the original patch grid. However, this can
be somewhat alleviated by manually setting the "Margin" value to large
values (5-10).

2. The original Martini model implements some of the bonds in ring
structures as constraints with LINCS. Since NAMD doesn't have this
feature, these constraints are implemented as normal bonds, but with
high force constant values. While this does not have any dramatic
effect on protein dynamics, it leads to regular crashes if large
timesteps (20 fs or higher) are used. As such, most of the times NAMD
Martini simulations are limited to 10 fs. Adding extra potentials such
as collective variables (one of NAMD's strongest features), steered MD
or even elastic networks (see below) makes things even worse.

3. The scripts preparing the structures in the RBCG tutorial DO NOT
implement elastic networks, other than the secondary structure
restraints from DSSP. If you want to use an elastic network with your
protein, you have to define it separately as an ExtraBonds file.
However, including extra these extra potentials increases the chances of
a crash to occur, even in 10 fs.

4. Features such as polarizable water, the new cholesterol model,
Martini DNA and Dry Martini use specialized GROMACS features (e.g.
virtual sites) and are not available in NAMD, unless you apply specific
changes to the source code. A search in the namd-l mailing list will
give you the answer for these.

All in all, I would advise against using vanilla Martini with NAMD, not
because GROMACS is necessarily better but because the force field itself
is tailored for GROMACS and its transfer to other codes is problematic
(this is not NAMD-specific, the LAMMPS implementation has many of the
aforementioned problems).

If you need to perform Coarse-Grained simulations with NAMD, I would
propose two alternatives (both, in fact, by the TCBG group): either the
original RBCG 2007 force field (which uses Martini for the lipids but
has its own model for proteins, fully compatible with NAMD) or the PACE
hybrid model, which uses Martini for the solvent/lipids and a United
Atom model for the proteins. The latter is very appealing, because it
combines the fast simulation of Coarse-Graining with a near atomic view
for the proteins.

Good luck,


PS. I feel like I've written these things way too many times for the
mailing list. At some point, the specifics of the implementation need
to be included in the User's Guide.

On 26-Jan-17 03:21, Panagiota Kyriakou wrote:
> Hello all,
> I followed the instructions from the RBCG tutorial and have created my
> system (a dimer embedded in a bilayer). Before I start running my
> simulation I have a few questions:
> 1) what version of martini parameters are the ones contained in the
> tutorial?
> 2) if someone has experience with CG would you think that Martini on
> NAMD will be significantly different than Martini on Gromacs
> Thank you,
> Giota
> Panagiota Kyriakou
> Ph.D. Candidate in Chemical Engineering
> Dept. of Chemical Engineering and Materials Science
> University of Minnesota

Fotis A. Baltoumas
Phd Candidate, Bioinformatics Postgraduate Programme
Department of Cell Biology and Biophysics
Faculty of Biology, University of Athens
Panepistimiopolis, Athens 157 01, GREECE
This email has been checked for viruses by Avast antivirus software.
** Visit for more info on majordomo