Application of Molecular Dynamics Simulation
- 1. Umbrella Sampling MD simulations for Retention Prediction in Peptide Reversed-Phage Liquid Chromatography Authors: Pablo M. Scrosati, Evelyn H. MacKay-Barr, and Lars Konermann. ACS Anal. Chem. Vol. 97 (1), 828-837(2025) Presenter: Kushal Raj Roy Date: 21 March, 2025
- 2. 2 Outlines: 2. Introduction to Molecular Dynamics. 3. Application of Umbrella Sampling for RPLC- Retention time prediction. 1. Introduction to Reversed-Phase Liquid Chromatography (RPLC).
- 3. 3 Introduction to Reversed-Phase Liquid Chromatography (RPLC) Alsaleh, Munirah, et al. 2019 • A widely-used analytical method for separating complex mixtures into individual chemical components. • Utilizes differences in analyte hydrophobicity to separate mixtures, often peptides or small biomolecules. Contains silica particles modified with hydrophobic alkyl chains:C4 (short), C8 (medium), C18 (long) chains.
- 4. 4 Outlines: 2. Introduction to Molecular Dynamics. 3. Application of Umbrella Sampling for RPLC- Retention time prediction. 1. Introduction to Reversed-Phase Liquid Chromatography (RPLC).
- 5. 5 Introduction to Molecular Dynamics 1930-2024
- 6. 6 Accuracy vs Efficiency trade-off for simulations
- 7. 7 Timescale of molecular events Bonfiglio et al., 2015.
- 8. 8 What is Molecular Dynamics? Classical Molecular Dynamics (Molecular Mechanics MD) is a simulation procedure consisting in computing the motions of atoms (individually or as part of a macromolecule), in a liquid or gas environment (solvent), using Newton’s motion Laws. Different from stochastic sampling methods used for molecular docking, here there is a time component, conecting all sampled conformations in an ordered series of events. Perilla, Juan R., et al. 2015
- 9. 9 Preparation for Molecular Dynamics You need to obtain the 3D structures of your input molcules (solute) • Experimentally-determined structures • Modeled structures; Fix the structures of your input molecules • No missing atoms; • No missing residues; • No additional atoms;
- 10. 10 Basic steps of Molecular Dynamics
- 11. 11 Software packages for MD There are many packages for Molecular Dynamics simulations: – GROMACS: Groningen Machine for Chemical Simulation (Free) – NAMD: Nanoscale Molecular Dynamics (Free) – OpenMM (Free) – AMBER: Assisted Model Building and Energy Refinment (Paid) – CHARMM: Chemistry at Harvard Macromolecular Mechanics (Paid) – Discovery Studio, part of the Schrodinger package (Paid) – MOE : Molecular Operating Environment (Paid)
- 12. 12 Force fields for MD There are many packages for Molecular Dynamics simulations: – GROMOS (developed for the GROMACS package; G54A6, G54A7, etc) – AMBER (developed for the AMBER package; AMBER14SB, etc) – CHARMM (developed for the CHARMM package; charmm36, etc) – OPLS (developed for the Schrodinger package; OPLS4, etc) – OpenFF (open source force field initiative)
- 13. 13 A Force Field can be defined as “a set of functions and parameters used for the interaction potential calculations in molecular mechanics MD” Each FF might have a different set of equations and parameters, to better reproduce aspects of the molecular behaviours observed in vivo/in vitro FF terms might account for bond stretching, angle changes, interactions, etc The 3 main categories of potential are: bonded, non-bonded, external fields. Introducing interaction potentials
- 14. 14 Therefore, the choice of the Force Field might be determined by the molecule you want to study, or the level of “simplification of the system.” Different levels of simplification include: • All atom model with explicit water • All atom model with implicit water • United atom model (e.g., hydrogens are merged to heavy-atoms; the system has pseudoatoms like CH, CH2 and CH3). • Coarse-grained model (e.g., entire amino acids might be represented as a single pseudoatom) Computational cost Introducing interaction potentials
- 15. 15 Levels of simplification of the system Singh et al.,
- 16. 16 Water models for MD
- 17. 17 Creation of the initial state
- 18. 18 Dahanayake et al., 2018 Energy minimization and solvation
- 19. 19 • Shape/volume of the box • Periodic Boundary Conditions • Equilibration of the system • Integration Time Step • Non-bonded interaction calculations After choosing the most adequate FF, water model and run the Energy Minimization, your system is ready for the simulation. But there some key MD parameters that you must consider: Parameters of the MD simulation
- 20. 20 The most common geometric shapes used in MD are related with a simulation strategy called Periodic Boundary Conditions (PBC). These geometrical shapes must enable the system to be replicated in each face of the box, in a periodic fashion. If a molecule leaves the central box to one side, it enters one of the replicated systems from the opposite side. But all systems are identical, so is like “reentering the same box”. Although we have a boundary in our system, we want that boundary to be “invisible” to the molecules in our system, so we are closer to the experimental conditions. Periodic Boundary Conditions
- 21. 21 Canonical ensemble (NVT) • Particle number N • Volume V • Temperature T • Total energy E • Pressure P Requires a thermostat, an algorithm that adds and removes energy to keep the temperature constant • Berendsen thermostat, v-rescale thermostat, etc… Equilibration of the system
- 22. 22 Equilibration of the system Isothermal-isobaric ensemble (NPT) • Particle number N • Pressure (P) • Temperature T • Total energy E • Volume V In addition to the thermostat, it requires a barostat, an algorithm that changes volume to keep the pressure constant • Parrinello-rahman barostat, Berendsen barostat, etc…
- 23. 23 Integration Time Step The calculations for the changes in the coordinates of all atoms of the system must be done in small steps, usually on the scale of femtoseconds (fs, 10-15 seconds) The succession of these steps (nsteps) will produce a series of “frames” (conformations) of the system, referred as a trajectory (i.e., a time series of the evolution of the system under defined conditions) The duration of these steps is called Integration time (dt). This value affects the efficiency and accuracy of the simulation. dt = 0.002 nsteps = 500.000 Time: 1 ns
- 24. 24 Non-bonded interaction calculations The calculation of non-bonded interactions (ionic and van der Waals interactions) also has an impact on the performance of the simulation While the bonded interactions increases proportionally with the number of atoms in the system, the number of non-bonded interactions increases as a function of the square of the number of the atoms Since the intensity of these interactions decrease quickly with the distance between the atoms, it is possible to define limits (cut-off) to which we want to calculate possible interactions.
- 25. 25 The production phase of the simulation After the energy minimization and the equilibration of the system, and having defined all previously mentioned parameters, we can start the so called “production phase” of the simulation This is the part of the simulation that we are interested in analyzing, so we can see how the system behaves over time The ideal length of the production phase will depend on what molecular changes we are trying to observe. Keep in mind the timescales of molecular events. For today standards, 50 ns is a short MD. Note that many of the initial parameters of the system, like the velocities for each atom, are not fixed values, but are obtained from a possible distribution of values. And these initial values can bias (a bit) the rest of the simulation. Therefore, a replicated simulation will produce somewhat different results (hopefully not so different). Recommended to run replicated experiments.
- 26. 26 Analysis of the MD simulation Hollingsworth et al., 2018
- 27. 27 Visual inspection of a trajectory (movie) Antunes et al., 2020
- 28. 28 Outlines: 2. Introduction to Molecular Dynamics. 3. Application of Umbrella Sampling for RPLC- Retention time prediction. 1. Introduction to Reversed-Phase Liquid Chromatography (RPLC).
- 29. 29 Preparation of Tryptic Myoglobin peptides
- 30. 30 Snapshots from microsecond MD Simulations
- 31. 31 Minimum distance between peptide and C18 atoms
- 32. 32 To capture the C18 binding behavior: Free Energy of Peptide Binding: ΔG°binding < 0 (strong retention), ΔG°binding > 0 (weak, or no retention) C18 Binding Behavior and Free Energy of Peptide Binding
- 33. 33 Microsecond MD Simulations: General Trends
- 34. 34 Enhance Sampling method -Umbrella sampling Róg, Tomasz, et al,
- 35. 35 Principles of Umbrella Sampling MD
- 36. 36 Umbrella Sampling for Retention Prediction
- 37. 37 G of peptides P1 P4 in a RPLC slit pore as − a function of distance from the stationary phase
- 38. 38 Peptide-C18 binding affinities ΔG from umbrella sampling vs experimental retention time
- 39. 39 Conclusion •Empirical methods: fast, accurate, but lack molecular details. •Standard MD: computationally costly, poor retention predictions (scrosati et al., 2023). •Umbrella sampling MD: accurate retention predictions via ΔG binding (scrosati et al., 2025). •Explicitly models hydrophobic, electrostatic, van der Waals, and hydrogen-bonding interactions.
- 40. 40 Future work •Utilize atomistic insights to optimize RPLC workflows. •Combine umbrella sampling data with existing predictive methods. •Expand umbrella sampling to longer peptides and intact proteins. •Explore in silico design of novel stationary phases, guiding
- 41. 41 What I could have done? Known retention times Features Secondary structures Physiochemical properties Free binding energy Output Protein behaviors Accuracy % solvent for SF Molecular Interaction GNN Architecture
Editor's Notes
- #31: dMIN represents the shortest distance between any peptide atom and any C18 atom. time points with peptide-C18 van der Waals contacts had dMIN ≤ 0.25 nm (peptidebound), while time points with a freely dissolved peptide had dMIN > 0.25 nm (peptidefree).
- #32: typical MD runs are too short for obtaining reliable ΔG°binding values in this way, causing eq 2 to suffer from problems analogous to those discussed above for f B
- #33: Can microsecond MD runs be used for retention prediction? In an RPLC column, only peptidefree is subject to convective transport, whereas net transport is stalled for peptidebound.8 In principle, MD-derived f B values should therefore correlate with retrntion behavior. Overall, the data in Figure 3 suggest that microsecond MD simulations are not a promising approach for peptide retention prediction.
- #35: the distance between peptide center of mass (DCOM) and the center layer of the SiO2 support. Schematic layout of umbrella sampling for probing peptide- C18 interactions. (A) n = 4 initial peptide positions. (B) Dashed lines represent umbrella potentials that restrain peptide COM motions. Bars represent biased distributions PBi(dCOM) for i = 1, ... n for each umbrella window. (C) WHAM-generated unbiased distribution P(dCOM), revealing peptide positional preferences within the pore. (D) Free energy G(dCOM) of the system. Red arrows indicate repulsive and attractive mean forces acting on the peptide in different regions of the pore. s a final step, WHAM converts P(dCOM) into a free energy profile G(dCOM) i
- #37: s noted, the peptide COM will always move toward the G(dCOM) minimum. Two tendencies are readily apparent: (1) At low %ACN, all G(dCOM) profiles have a minimum at the C18/mobile phase interface (Figure 6A− H), promoting peptide binding to the C18 chains. At higher % ACN, the profiles gradually morph into hyperbolic shapes, implying a preference for freely dissolved peptides close to the pore center
- #38: plots of ΔGbinding vs experimental retention time showed linear relationships (Figure 7). The best correlations (R2 ≥ 0.94) were found for water, 10%, and 30% ACN (Figure 7A−C).