A new fat node has been added to the cluster. This is based on an Intel Core i7-13700KF (1700/3.4 GHz/30 MB), an NVIDIA MSI RTX 4070 Ti VENTUS 3X 12Gb OC 3, 8GB DDR4@3200 MHz, and two 2TB disks in a RAID1 array.

Initial tests confirmed that this is our new fastest machine, delivering with gromacs and HMR ~3930 ns/day on the FipWW 8,000 atom/5fs system. As with the other machines, with smaller systems the performance flattens at approximately ~4.1 μs/day (again with gromacs & HMR).

The joy of WSL2

 And just like that, there is no reason to fight with building native windows executables any longer. Install WSL2 on win11, and be done with it : all Linux graphical apps work out of the box, and at near native speeds. The screenshot below is from a molecular dynamics simulation analysis with carma / grcarma / stride / rasmol / evince / ..., all visibly working. 

Martini & Gromacs & VMD & Bendix (and ROP)

 Have been playing with the Martini force field for some time now. Still digging. Here is a short movie of a ROP mutant using Martini & Gromacs, and visualized with VMD & Bendix.

Just came out ...

«A molecular dynamics simulation study on the propensity of Asn-Gly-containing heptapeptides towards β-turn structures: Comparison with ab initio quantum mechanical calculations.»

https://doi.org/10.1371/journal.pone.0243429

New node added to the cluster

A new fat node has been added to the cluster. This is based on an Intel i9-9900K, an RTX2070S, 8GB DDR4@3000 MHz, and a 2TB disk.

Initial tests confirmed that this is our new fastest machine, delivering ~575 ns/day on the standard 10,000 atom/4fs benchmark. As with the other i9, with smaller systems the performance flattens at approximately ~700ns/day with a 2.5fs step.

Illustrating transitions between conformers using circos plots

A new tutorial  is available for preparing a circos plot illustrating the transitions between stable conformers identified from e.g. a folding molecular dynamics simulation.

OpenDX rendering of folding landscapes

Folding landscape (from dPCA) of the M2TM peptide in TFE, rendered with OpenDX. Too much red probably.

Molecular simulation of peptides review

Just came out : "Molecular simulation of peptides coming of age: Accurate prediction of folding, dynamics and structures"

http://doi.org/10.1016/j.abb.2019.01.033

αLa : add 15SB to the family

The same image as before, after adding results from the 15SB force field :

A definite improvement over both 12SB and 14SB.

gp41 paper

Just came out : "Folding Molecular Dynamics Simulation of a gp41-Derived Peptide Reconcile Divergent Structure Determinations", ACS Omega, 3, 14746-14754 :

http://doi.org/10.1021/acsomega.8b01579

Just came out :

«Folding Simulations of a Nuclear Receptor Box-Containing Peptide Demonstrate the Structural Persistence of the LxxLL Motif Even in the Absence of Its Cognate Receptor»

http://dx.doi.org/10.1021/acs.jpcb.7b10292

Amber 14SB vs 99SB-STAR-ILDN [2]

Have been looking for a mostly disordered peptide (with NMR data available) for which the two force fields would demonstrate detectably different secondary structure preferences. I think I found one :

The upper graph is from 99SB-STAR-ILDN, the lower from 14SB. The two simulations were 24 μs each, both using adaptive tempering (280K-380K). Comparison between observed and calculated  NOEs plus chemical shifts should suffice. Given that this is a mostly disordered peptide ,we should probably also compare the computationally expected vs experimentally observed number of NOEs.

Amber 14SB vs 99SB-STAR-ILDN : αLa peptide

Weblogo representations of secondary structure preferences for human α-Lactalbumin 101-111 peptide with the AMBER ff14SB plus a whole lot of other AMBER force fields. The 14SB simulation was 3 μs, all other 2 μs.

On the way from 12SB to 14SB the α-helical preference was significantly reduced, but the mainly 3₁₀-helical nature of this peptide can not be faithfully reproduced. At least for the time, AMBER99SB-STAR-ILDN still looks like the best force field for this peptide.

Amber99SB-STAR-ILDN : tri-alanine

Added one extra diagram to Figure 3 of this paper. This is a 2.6 μs simulation with a 4 fs timestep (HMR). Nearly identical with AMBER 14SB (?).

New fat compute node arrived

New fat compute node arrived. It contains an i7-6800K@3.4GHz, an nvidia GTX-1070, 16 GBytes of physical memory and two WD Red 1TB disks. After a few problems with setting-up the disks as RAID0 (using Intel's rapid storage), a minimal CentOS 7.3 was installed and the nvidia drivers compiled and installed.

First tests with NAMD 2.12 were impressive : on a small peptide problem with 4400 atoms, the new box delivered ~330 nanoseconds per day, a speed-up by a factor of 10 compared with NAMD 2.9 running on a node containing an 8-core AMD FX-8150 plus a GTX 570. To make the comparison hardware-dependent-only, we should test with NAMD 2.9 on the new box (testing new NAMD on the old box won't do because support for fermi GPUs was dropped with NAMD 2.12).

On the same peptide system (4300 atoms) and with a 2.5 fs time step, 470 ns/day were recorded. This reached an outstanding 550 ns/day for a smaller (3300 atoms) system. Think about it again : more than 1 μs in two days. Two days.

Update : It seems that with this hardware and NAMD 2.12 it is necessary to disable off-loading FFT to the GPU by using the 'usePMECUDA no' configuration option.

New paper from the group just came-out :

A new method for quantifying convergence of biomolecular simulations.

We have developed a new probabilistic method for quantifying convergence of molecular dynamics simulations. The essence of our method is the following : We treat the molecular dynamics trajectory as a finite sample of “molecular species” (clusters of similar structures) taken from an underlying distribution containing an unknown number of such molecular species. The observed frequencies of molecular species in the sample are calculated, and the Good-Turing formalism is applied to these frequencies allowing us to estimate the total probability of unseen (i.e. as yet unobserved) species. The result is the answer to the following question: "What is the probability that a molecular configuration with an RMSD (from all other already observed configurations) higher than a given threshold has not actually been observed?".

A paper containing a full description of the method is available from JCIM.

A program implementing the method is available via the github repository.

Force-field dependent secondary structure preferences

Weblogo representations of the per-residue secondary structure preferences (as produced by STRIDE) for folding simulations of the α-Lactalbumin-derived peptide studied in this paper. Results from seven force fields are shown. The experimental NMR results indicate a mostly 310-helical N-terminal part (residues 3-6) with an occupancy of ~50%, and a completely disordered C-terminus. The symbols in the weblogo diagrams are G => 310 helix, H => α helix, T => turn, C => random coil, E => extended. The force fields are CHARMM22, OPLSaa, AMBER ff12SB, AMBER ff99SB, and three variants of AMBER ff99SB (99SB-ILDN-NMR, 99SB-ILDN, 99SB-STAR-ILDN). It does look like a clear take-home message is present in these diagrams...

grcarma on the cover of JCC

The program grcarma written by Panagiotis Koukos made the cover story of the latest issue of the Journal of Computational Chemistry. Well done Panagiotis !

Homepage update

Added Kontopoulos' and Koukos' papers. Screenshots and artistic renderings follow ... :-)