Inverse Monte Carlo Method for Determination of Effective Potentials for Coarse-Grained Models IPAM Workshop "Multiscale Modeling in Soft Matter and Biophysics September 26-30, 2005 Alexander Lyubartsev (sasha@physc.su.se) Division of Physical Chemistry Arrhenius Lab., Stockhom University Outline 1. Introduction why do we need multiscale coarse-grained modeling 2. Inverse Monte Carlo Method how to build effective potentials for coarse-grained models 3 Effective solvent-mediated potentials ion-ion and ion-DNA 4 Coarse-grained lipid model large-scale simulations of lipid assemblies Why do we need coarse-grained modeling? Ion density profile ( M/l ) a) polyelectrolyte problem: ions around DNA (+2) MC (+2) PB (+1) MC (+1) PB (-1) MC (-1) PB 1 0.1 10 20 30 40 r (Å) Atomistic MD: not really possible to sample distances 3040 Å from DNA Primitive model (MC) - how good it is? b) Lipid bilayer in water Lipid (DMPC) All-atom MD: 1 lipid - more than 100 atoms (DMPC -118, DPPC - 130) ''minimal'' piece of bilayer: 6x6x2 = 72 lipids add at least 20 water molecules per lipid ⇒ about 13000 atoms (the picture above contains about 50000 atoms) A good object to waste CPU time.... Multiscale approach All-atomic model Full information MD simulation (but limited scale) Coarse-graining – simplified model Effective potentials for selected sites Reconstruct potentials (inverse Monte Carlo) RDFs for selected degrees of freedom Increase scale Effective potentials Properties on a larger Simulation of coarse length/time scale grained model (MD,MC,BD,DPD...) Inverse Monte Carlo Model Interaction potential direct inverse Properties Radial distribution functions •Effective potentials for coarse-grained models from "lower level" simulations (atomistic coarse grained; CPMD atomistic) •Reconstruct interaction potential from experimental RDF •An interesting theoretical problem The method (A.Lyubartsev and A.Laaksonen, Phys.Rev.A.,52,3730 (1995)) Consider Hamiltonian with pair interaction: V H V (rij ) i, j Make “grid approximation”: H V S Hamiltonian can be rewritten as: | | | | | | | =1,…,M Where V=V(Rcut/M) - potential within -interval, S - number of particle’s pairs with distance between them within -interval Note: S is an estimator of RDF: 1 V g (r ) S 2 2 4r r N / 2 Rcut Set of V , =1,…,M Space of Hamiltonians direct {<S>} {V} inverse In the vicinity of an arbitrary point in the space of Hamiltonians one can write: S S V V O(V 2 ) where dqS (q) exp V S (q) S S S S S V V dq exp V S (q) = 1/kT q r1 ,..., rN Choose trial values V(0) Direct MC Calculate <S> (n) and differences <S>(n) = <S >(n) - S* Solve linear equations system Obtain V(n) New potential: V(n+1) =V(n) +V(n) An analogueS Newton method S S(V) V S* V * V1 V0 V Repeat until convergence Algorithm: Initial approximation: mean force potential V(0) =-kTln(g*(r)) Some comments • Solution of the inverse problem is unique for pair potentials (with exception of an additive constant) gik(r) Vik(r)+const • There exist a simpler scheme to correct the potential: V(n+1)(r) = V(n)(r) + kT ln(g(n)(r)/gref(r)) (A.K.Soper, Chem.Phys.Lett, 202, 295 (1996)) Its convergence is however slower and may not work in multicomponent case • The precision of the inverse procedure can be defined by analysing eigen values and eigen vectors of the matrix S V Effective solvent-mediated potentials. Two levels of simulation of ionic, polymer or other solutions: 1) All-atom simulations (MD) with explicit water. 10000 atoms - box size ~ 40 Å 2) Continuum solvent, solutes - some effective potential, for example, ions - hard spheres interacting by Coulombic potential with suitable e. Ion radius - adjustable parameter (so called "primitive electrolyte model") The idea is to build effective solvent-mediated potential, which, maintaining simplicity of (2), takes into account molecular structure of the solvent A. Effective solvent-mediated potentials between Na+ and Cl- ions Reference MD simulations: H2O Na+ Cl- flexible SPC model (K.Toukan, A.Rahman, Phys.Rev.B31, 2643 (1985) s=2.35Å, e=0.544 kJ/M s=4.4Å, e=0.42 kJ/M (D.E.Smith, L.X.Dang, J.Chem.Phys., 100, 3757 (1994) Double time step algorithm, with short time step 0.2fs and long time step 2fs, was used NPT-ensemble, T=300K, P=1atm, Ion-ion effective potentials Ion-ion RDFs 4,0 6 3,0 RDF 2,5 2,0 1,5 1,0 0,5 0,0 NaCl, L=39Å NaCl, L=24Å NaNa, L=39Å NaNa, L=24Å ClCl, L=39Å ClCl, L=24Å Prim. model 4 Eff. Potential / kT NaCl, L=39Å NaCl, L=24Å NaNa, L=39Å NaNa, L=24Å ClCl, L=39Å ClCl, L=24Å 3,5 2 0 -2 5 r (Å) 10 5 r (Å) 10 15 NaCl osmotic and activity coefficients Solvent-mediated effective potentials were applied to calculate osmotic and activity coefficients of Na+ and Cl- ions in the whole concentration range. MC simulations are carried out for 200 ion pairs using effective potentials Osmotic coefficient: osm. coef activity coef 1.3 1.2 1 F / c P V osm kT c T NkT Activity coefficient: ex exp kT 1.1 1.0 0.9 0.8 0.7 0.6 1E-3 0.01 0.1 1 Concentration (M) Lines are calculated values and points are experimental data B. Ion-DNA effective solvent-mediated potentials Molecular dynamics: • One turn of DNA (dATGCAGTCAG): 635 atoms, CHARMM force field (A.D.MacKerell, J.Wiorkiewicz-Kuchera, M.Karplus, JACS, 117, 11946 (1995)) • flexible SPC water model + ions: Run 1 2 3 4 5 No. of H2O 500 500 1050 1050 500 Counterions 20 Li+ 20 Na+ 30 Na+ 30 K+ 20 Cs+ Coions - - 10 Cl- 10 Cl- - Simulation time (ns) 2.5 2 2.5 2.5 1.5 All-atom model: Coarse-grained model Na+ Ion - DNA effective potentials Ion - P 6 + Li 6 + + Na 4 + K + Cs 0 -2 -4 + Na Eff. potential / kT + Li + Na * 2 + Li Na + 4 Na * + K* + Cs 2 0 Eff. potential / kT 6 Eff. potential / kT Ion - C4’(sugar) Ion - C4 (base) + 4 Na * + K* + Cs 2 0 -6 2 4 6 8 r (Å) 10 12 14 16 2 4 6 8 r (Å) 10 12 14 16 -2 2 4 6 8 r (Å) 10 12 14 16 MC simulation: a bigger DNA fragment (3 turns) in a box 100x100x102Å, ions interacting by effective solvent-mediated potentials; no explicit water. These are results for the density profile and integral charge 1.0 + Li 1 0.8 + NA * Integral charge Density profile (M/l) + Na + K* + Cs PB 0.1 + 0.6 Li + Na + Na * 0.4 + K* + Cs PB 0.2 0.01 0 10 20 30 r (Å) 40 50 0.0 0 10 20 30 r (Å) 40 50 Relative binding affinities of ions The order of relative binding affinities of alkali counterions to DNA, defined by MC simulation with effective potentials, is: Cs+ > Li+ > Na+ > K+ The binding order was defined also in a number of experimental works: •P.D.Ross, R.L.Scruggs, Biopolymers, 2, 89 (1964) ; Electrophoresis: Li+>Na+>K+ •U.P.Strauss, C.Helfgott, H.Pink, J.Phys.Chem.,71,2550 (1967); Donnan equilibrium: Li+>Na+>K+ •S.Hallon et al, Biochemistry, 14, 1648 (1975); Circular dichroism Cs+>Li+>K+>Na+ •P.Anderson, W.Bauer, Biochemistry, 17, 594 (1978), DNA supercoiling Cs+>Li+>K+>Na+ •M.L.Bleam, C.F.Anderson, M.T.Record, Proc. Natl.Acad.Sci USA,77,3085 (1980), NMR: •I.A.Kuznetsov et al, Reactive Polymers, 3, 37 (1984), Ion exchange Cs+>Li+>K+>Na+ Li+>K+Na+ Qualitative agreement with results of experiments of very different nature. Coarse-grained lipid model All-atom model 118 atoms Coarse-grained model 10 sites We need interaction potential for the coarse-grained model ! Use IMC and RDFs from atomistic MD. All-atomic molecular dynamics All-atomic MD simulation was carried out: 16 lipid molecules (DMPC) dissolved in 1600 waters (6688 atoms) Box size: 40x40x40 Å ● Initial state - randomly dissolved RDFs calculated during 12 ns after 2 ns equilibration ● ● Force field: CHARMM 27, water - flexible SPC ● T=313 K MD snapshot 16 DMPC lipids 1600 H2O R D F calculations N P 4 different groups -> 10 pairs 10 RDFs and eff. intermolecular potentials + 4 bond potentials CO C 7 6 solid - 16 lipids dashed - 64 lipids NP 6 5 4 4 PP 3 RDF RDF 5 NN 2 1 0 3 CC 2 PC 1 5 10 r(A) 15 20 0 NC 5 10 r(A) 15 20 Inverse MC simulations: Purpose: find effective potentials which, for the coarse grained model, reproduce the same RDFs as the all-atomic model Intramolecular potentials: Bonded: from distance distribution between the atoms. Non-bonded - the same as intermolecular Total: 14 effective potentials Inverse MC - the box of the same size; the same number of lipids as in the corresponding MD; no solvent: charges +1 and -1 on "N" and "P" + dielectric constant e=70 (best fit to NN, NP and PP -potentials) Effective potentials: 20 20 N - CO N - CH P - CO P - CH 15 Veff (kJ/M) Veff (kJ/M) NN 10 PP 0 0 5 0 NP -10 10 5 10 15 -5 20 r(A) 5 10 15 r(A) 20 15 N-P P - CO CO - CH CH - CH 20 Veff (kJ/M) Veff (kJ/M) 30 CO - CO CO - CH CH - CH 0 -10 r (A) Bond potentials 20 10 10 5 20 10 0 3 4 5 r(A) 6 7 8 Coarse-grained simulations ● Monte Carlo ● Molecular Dynamics Forces - from the potentials difference in the neigbouring grid points Solvent is not present explicitly - MD may be considered only as another way to generate canonical ensemble Time step 10-14 s + thermostat ● ● ● Nose-Hoover Local (Lowe-Andersen) Langevine 3 cases : a periodic bilayer a finite piece of bilayer random initial state Equivalent all-atom simulations would correspond ~ 106-108 atoms Infinite bilayer Z Periodic box Density distribution Coarse-grained MC (392 lipids) All-atom MD (98 lipids) A sheet of bilayer The same initial state, but in a large simulation box: End of simulation: 109 - 1010 MC steps: View from the side View from the top (discoid shape) Vesicle formation Start from a square plain piece of membrane, 325x325 Å, 3592 lipids: cut plane Membrane self-assembly MD simulation of 392 CG lipids with Lowe-Andersen thermostat http://www.fos.su.se/physical/sasha/lipids Conclusions 1. The multiscale approach based on the inversion of radial distribution functions provides a straightforward way to build effective potentials for coarse-grained models 2. Examples of ionic solutions, ion-DNA interactions, lipid membranes show that effective potentials, derived exclusively from the atomistic model, provide realistic description for the coarse-grained model 3. Coarse-grained effective potentials may be plugged in into MC, MD, Brownian dynamics, DPD and used for simulation on larger length- and time scale Acknowledgements Aatto Laaksonen Martin Dahlberg Carl-Johan Högberg