Glossary of computational chemistry terms

Molecular dynamics

• Molecular dynamics (MD). A computer simulation that follows how every atom of a system moves over time by solving Newton's equations of motion, with the forces taken from a force field. It is used to watch how a protein, a ligand or a complex behaves in water. See what is molecular dynamics.
• Force field. The set of equations and parameters that gives the potential energy of a system from the positions of its atoms, with terms for bonds, angles, torsions, van der Waals contacts and electrostatics. AMBER, CHARMM and OPLS are the common families.
• AMBER ff14SB and ff19SB. Protein force fields of the AMBER family. ff19SB is the newer one and is meant to be used with the OPC water model.
• CHARMM36m. A protein force field of the CHARMM family that also describes flexible and disordered regions well. It is used with a modified TIP3P water model.
• OPLS-AA/M. An all atom force field of the OPLS family that was refined for proteins.
• GAFF2. The General AMBER Force Field, version 2, which gives parameters for drug like small molecules. It is usually combined with AM1-BCC charges.
• AM1-BCC charges. A fast way to give a ligand partial atomic charges, by AM1 charges plus bond charge corrections that imitate RESP charges.
• RESP charges. Restrained electrostatic potential charges. Atomic charges fitted to the electrostatic potential of a quantum calculation, with restraints that keep them chemically sensible. A common choice for ligand parameters.
• Time step. The interval between two steps of a simulation, usually 2 femtoseconds when bonds to hydrogen are constrained and 4 femtoseconds with hydrogen mass repartitioning.
• Hydrogen mass repartitioning (HMR). Moves mass from heavy atoms to the hydrogens bonded to them, which lets the time step grow to 4 femtoseconds without making the simulation unstable.
• SHAKE and LINCS. Algorithms that keep bonds to hydrogen at a fixed length, so that a 2 femtosecond time step is possible.
• Trajectory. The saved record of atom positions at regular intervals during a simulation. It is what you analyze afterwards.
• Periodic boundary conditions. The simulation box is repeated in every direction, so an atom that leaves one side comes back in on the opposite side and there is no surface in the system.
• Simulation box. The cell that holds the solute and the water. Cubic, truncated octahedron and rhombic dodecahedron boxes are common, and the last two need fewer water molecules.
• Explicit solvent. Every water molecule is simulated as its own set of atoms. It is more accurate and much more costly than implicit solvent.
• Implicit solvent. Water is replaced by a continuous medium with the dielectric properties of water. It is fast and is used for quick estimates such as MM/GBSA.
• Water model. The force field of the water molecule. TIP3P, TIP4P-Ew, SPC/E and OPC are common, and the one you pick should be the one your protein force field was made for.
• Counterions and ionic strength. Sodium and chloride ions are added to neutralize the system and to reach about 0.15 M salt, which imitates physiological conditions.
• Protonation state. Whether an acidic or basic group carries a hydrogen at the chosen pH. Histidines and catalytic residues need special attention before a simulation or a docking run.
• Energy minimization. Moves atoms downhill in energy to remove clashes before dynamics starts. Steepest descent, sometimes followed by conjugate gradient, is typical.
• Equilibration. A short run, often with restraints on the solute at first, that brings temperature, pressure and density to the target values before the production run.
• NVT ensemble. Constant number of atoms, volume and temperature. It is used to heat the system.
• NPT ensemble. Constant number of atoms, pressure and temperature. It is used to settle the density and for most production runs.
• Production run. The long simulation after equilibration whose trajectory you analyze. For a protein ligand complex, 100 ns with several replicates is a common standard.
• Thermostat. The method that keeps the temperature constant, for example Nosé-Hoover, Langevin or velocity rescaling.
• Barostat. The method that keeps the pressure constant, for example Parrinello-Rahman, Berendsen or Monte Carlo.
• Particle Mesh Ewald (PME). The standard way to calculate long range electrostatic forces accurately and fast in a periodic box.
• Cutoff. The distance, often 1.0 to 1.2 nm, beyond which short range van der Waals and electrostatic interactions are not calculated directly.
• Lennard-Jones potential. The equation for van der Waals interaction, with weak attraction at distance and strong repulsion at close range.
• Replicate. An independent repeat of a simulation from different starting velocities. Three replicates show whether a result can be reproduced.
• Convergence. The point where a measured property stops drifting and fluctuates around a stable average. A property that has not converged cannot be trusted.
• RMSD. Root mean square deviation, the average distance in Å between the atoms of a structure and a reference, usually after fitting. A backbone RMSD that levels off suggests a stable protein, and a ligand RMSD within about 2 to 3 Å suggests the pose holds.
• RMSF. Root mean square fluctuation of each residue around its average position. Peaks show flexible loops and chain ends, and the values can be compared with crystallographic B factors.
• Radius of gyration (Rg). A measure of how compact a protein is. A steady value means the fold stays compact and a rising value means it is opening up.
• SASA. Solvent accessible surface area, the part of a molecule that water can touch. It falls when a ligand buries part of a surface.
• Hydrogen bond occupancy. The percentage of frames in which a given hydrogen bond exists. Interactions with high occupancy are the ones that hold the ligand.
• Principal component analysis (PCA). Finds the few collective motions that explain most of the movement in a trajectory. The first two or three components are plotted in 2D or 3D.
• Free energy landscape (FEL). A map of how often the system visits each conformation, drawn from PCA components or other coordinates. Low free energy regions are the favoured states.
• Dynamic cross correlation matrix (DCCM). Shows which residues move together and which move in opposite directions during the simulation.
• DSSP. An algorithm that assigns secondary structure, such as helix and strand, to every residue in every frame, so that changes in structure can be followed.
• Clustering. Groups similar frames of a trajectory and gives a representative structure for each group.
• MM/PBSA. Estimates a binding free energy from snapshots of a trajectory by adding the molecular mechanics energy to a Poisson-Boltzmann polar solvation term and a surface area term. It ranks ligands against each other but is not an absolute binding free energy.
• MM/GBSA. The same idea as MM/PBSA with the faster generalized Born model for the polar solvation term.
• Per residue energy decomposition. Splits an MM/PBSA or MM/GBSA energy into the contribution of each residue, which shows which residues hold the ligand.
• Entropy term. The part of a binding free energy that comes from the loss of freedom on binding. Normal mode and quasi harmonic estimates are noisy, so many studies leave it out and say so.
• Free energy perturbation (FEP). A rigorous method that changes one ligand into another in many small steps to get the difference in binding free energy. It is more accurate and much more expensive than MM/PBSA.
• Steered MD. Pulls a ligand out of its pocket with a moving restraint to see how strongly it is held.
• Umbrella sampling. Runs many windows along a path, each with its own restraint, and combines them into a potential of mean force.
• Potential of mean force (PMF). The free energy profile along a chosen coordinate, such as the distance of a ligand from its pocket.
• Enhanced sampling. Methods such as metadynamics and replica exchange that help a simulation cross energy barriers that plain MD would need far too long to cross.
• Alanine scanning. Replaces residues with alanine, in the experiment or in a calculation, to see which side chains matter most for binding.
• Allosteric pocket. A binding site away from the active site whose occupation changes the activity of the protein.
• Cryptic pocket. A pocket that cannot be seen in the crystal structure and opens only during motion, which is why MD is used to find it.
• GROMACS, AMBER, OpenMM and NAMD. The main MD programs. GROMACS and OpenMM are open source, the fastest GPU engine of AMBER is licensed and NAMD is free for non commercial use.
• MDAnalysis and VMD. Tools to read trajectories, calculate properties and visualize them.
• CHARMM-GUI. A web tool that builds the input files for several MD programs.

Molecular docking and virtual screening

• Molecular docking. Predicts how a small molecule binds to a protein and scores how well it fits. See how to do molecular docking.
• Receptor. The protein, or other macromolecule, that the ligand is docked into.
• Ligand. The small molecule that is docked, for example a drug candidate.
• Binding site. The pocket on the protein where the ligand binds. It is often the active site of an enzyme.
• Grid box. The search space of a docking run, given by a centre and a size in Å. The ligand must stay inside it. A box that is only a little larger than the pocket gives the best results.
• Blind docking. Docking with a box that covers the whole protein, used when the binding site is not known. It is less reliable and slower than targeted docking.
• Targeted docking. Docking into a known pocket, with the grid box centred on a co crystal ligand or on key residues.
• Co crystal ligand. The ligand that sits in the pocket of a crystal structure. It defines the binding site and is the reference for redocking.
• Redocking. Docking the co crystal ligand back into its own protein to test the protocol. An RMSD below 2 Å to the crystal pose counts as a successful validation.
• Cross docking. Docking a ligand into a structure that was solved with a different ligand, which tests how well a protocol copes with a protein that has moved.
• Pose. One predicted position and orientation of the ligand in the pocket. A run reports several poses ranked by score.
• Scoring function. The equation that estimates how well a pose binds. Vina uses an empirical function and GNINA adds a neural network.
• Binding affinity. How strongly a ligand binds, given as ΔG in kcal/mol or as Ki, Kd or IC50. A more negative ΔG means tighter binding.
• Docking score. The estimated binding energy of a pose in kcal/mol. Scores are useful to rank ligands against the same protein and are not exact binding energies. Compare against a known binder instead of trusting a single number.
• Exhaustiveness. How much effort Vina spends on the search. The default is 8, and higher values such as 16 or 32 give more reproducible poses at the cost of time.
• Number of modes. How many poses are written for each ligand. The default is 9.
• AutoDock Vina. A fast, widely used open source docking program with an empirical scoring function. ChemServer runs version 1.2 on CPU.
• AutoDock 4. The older docking program of the AutoDock family that uses pre calculated grid maps and a force field based scoring function.
• Vina-GPU. A version of Vina that runs the search on a graphics card and is much faster on large libraries.
• GNINA. A docking program built on Vina that rescores poses with convolutional neural networks.
• CNN score. The score from the convolutional neural network of GNINA. CNN pose score tells how likely a pose is correct and CNN affinity estimates the binding strength.
• PDBQT. The file format of AutoDock and Vina. It holds atom coordinates, partial charges, AutoDock atom types and the torsion tree. It holds no bond orders.
• Gasteiger charges. Quick partial charges calculated from electronegativity and used in PDBQT files.
• Rotatable bonds and torsions. The bonds around which a ligand can turn during docking. More torsions make the search harder.
• Flexible docking. Docking in which chosen side chains of the receptor are allowed to move as well as the ligand.
• Induced fit. The change of a protein shape when a ligand binds. Rigid docking cannot show it, MD and flexible docking can.
• Virtual screening. Docking a large library of molecules to find those most likely to bind a target, before any lab work. See virtual screening on ChemServer.
• Hit. A molecule from a screen that shows promising activity or a good score and deserves a closer look.
• Enrichment. How many true actives a screen finds in its top ranks compared with random picking. It is reported as an enrichment factor or an ROC curve.
• Decoys. Molecules assumed to be inactive, with properties matched to the actives, used to test whether a docking protocol can tell binders from non binders.
• Protein preparation. Fixing a structure before docking by adding missing atoms and hydrogens, setting protonation states and removing water and extra ligands. See protein repair.
• Tautomer. One of several forms of a molecule that differ in where a hydrogen sits. The form you dock can change the result.
• Crystal waters. Water molecules in a crystal structure. Most docking runs remove them, but a water that bridges the ligand and the protein can matter.
• Interaction fingerprint. A bit string that records which residues a pose contacts and how, used to compare poses and to read results. ChemServer draws these with ProLIF.
• Hydrogen bond. An attraction between a donor with a hydrogen and an acceptor atom, usually about 2.7 to 3.3 Å between the heavy atoms. It is the most important directional interaction in drug binding.
• Salt bridge. A hydrogen bond between groups of opposite charge, for example a carboxylate and an ammonium.
• Pi stacking. The attraction between aromatic rings, face to face or edge to face.
• Cation pi interaction. The attraction between a positive charge, such as a lysine or arginine side chain, and the face of an aromatic ring.
• Halogen bond. An attraction between a halogen atom such as chlorine, bromine or iodine and a Lewis base such as a carbonyl oxygen.
• Hydrophobic contact. Contact between non polar parts of the ligand and the protein, a major source of binding energy that docking drawings show as arcs or lines.
• Ligand efficiency. Binding energy divided by the number of heavy atoms. It helps to compare small and large ligands fairly.
• Pose validation. Checking a predicted pose against a known structure by RMSD, or checking its geometry for clashes and strain.
• Homology model and AlphaFold structure. A predicted protein structure. It can be used for docking, but its side chains in the pocket should be checked, because small errors change the result.

DFT and quantum chemistry

• Density functional theory (DFT). A quantum mechanical method that finds the energy and electron density of a molecule from the density itself. It is the most widely used method for molecules of drug size. See how to do a DFT calculation.
• Functional. The approximation for the exchange and correlation energy in DFT. Its choice decides the quality of the result.
• B3LYP. A very popular hybrid functional. It is reliable for geometries and orbital energies and should be used with a dispersion correction.
• Dispersion correction (D3, BJ damping). An added term that describes the weak attraction between atoms that many functionals miss. B3LYP-D3(BJ) is a standard combination.
• M06-2X and ωB97X-D. Functionals that handle non covalent interactions and reaction energies better than plain B3LYP.
• Basis set. The set of functions used to describe orbitals. Bigger sets are more accurate and slower. def2-SVP is common for geometries and def2-TZVP for final energies.
• 6-31G(d) and 6-311++G(d,p). Pople basis sets. The d and p add polarization functions and the plus signs add diffuse functions that matter for anions.
• Geometry optimization. Moves the atoms until the energy is at a minimum. It must be followed by a frequency calculation.
• Frequency calculation. Gives vibrational frequencies, IR and Raman spectra and thermal corrections. A minimum has no imaginary frequency.
• Imaginary frequency. A frequency that comes out negative. One imaginary frequency marks a transition state and more than one in a minimum means the structure is not at a true minimum.
• Transition state. The highest point on the lowest energy path between reactants and products. It has exactly one imaginary frequency.
• Intrinsic reaction coordinate (IRC). The path that runs downhill from a transition state to reactants and products, used to prove that a transition state connects the right minima.
• Nudged elastic band (NEB). A method that finds a reaction path and a good guess for the transition state between two structures.
• Single point energy. The energy of a fixed geometry, often at a higher level than the one used for the optimization.
• Implicit solvation model (SMD, CPCM). Treats the solvent as a continuum around the molecule. SMD is a common choice for water, DMSO and other solvents.
• HOMO and LUMO. The highest occupied and the lowest unoccupied molecular orbital. They are the frontier orbitals that take part in electron donation and acceptance.
• HOMO LUMO gap. The energy difference between the HOMO and the LUMO. A small gap suggests a more reactive and more polarizable molecule and a large gap suggests a more stable one.
• Chemical hardness and softness. Hardness is half of the HOMO LUMO gap. Soft molecules are more polarizable and more reactive.
• Electrophilicity index. A descriptor built from the chemical potential and the hardness that ranks how strongly a molecule accepts electrons.
• Fukui function. Shows which atoms of a molecule are most likely to accept or give electrons, and so where a nucleophile or an electrophile will attack.
• Molecular electrostatic potential (MEP). The electrostatic potential on the surface of a molecule. Red regions are electron rich and blue regions are electron poor.
• Natural bond orbital (NBO) analysis. Describes bonds, lone pairs and charge transfer in terms of localized orbitals and gives natural charges.
• E(2) stabilization energy. The second order perturbation energy in NBO analysis that measures how much a donor orbital stabilizes an acceptor orbital, for example a lone pair into an antibonding orbital.
• Atomic charges (Mulliken, Hirshfeld, ESP). Ways to split the electron density among atoms. Mulliken charges depend strongly on the basis set and ESP charges reproduce the electrostatic potential.
• Time dependent DFT (TD-DFT). Calculates excited states, which gives UV Vis spectra, fluorescence and electronic circular dichroism.
• UV Vis spectrum. The absorption of light by electronic transitions. It is calculated with TD-DFT and compared with the measured spectrum.
• IR and Raman spectra. Vibrational spectra from the frequency calculation. Calculated frequencies are usually scaled by a factor near 0.96 to match the experiment.
• NMR shielding (GIAO). Calculates the magnetic shielding of each nucleus, which is converted to chemical shifts by a reference such as TMS.
• Electronic circular dichroism (ECD). The difference in absorption of left and right circularly polarized light. Calculated ECD is used to assign absolute configuration.
• Density of states (DOS). A plot of how many orbitals lie at each energy, which shows the contribution of fragments to the frontier orbitals.
• Charge and spin multiplicity. The two numbers every quantum input needs. A closed shell neutral molecule has charge 0 and multiplicity 1.
• Zero point energy and thermal corrections. Corrections from vibrations that turn an electronic energy into an enthalpy or a Gibbs free energy.
• Gibbs free energy. The energy that decides whether a process is spontaneous. From DFT it is the electronic energy plus the thermal correction at a chosen temperature.
• Relaxed scan. Changes one bond, angle or torsion step by step and optimizes everything else at each step, giving a rotation or reaction profile.
• E and Z isomers, syn and anti rotamers. E and Z isomers differ in the arrangement around a double bond, and syn and anti rotamers differ in the rotation around a single bond such as the amide bond of an acylhydrazone. DFT energies show which form is favoured.
• ORCA. A quantum chemistry program that is free for academic use and widely used for DFT, TD-DFT and spectra. ChemServer DFT work uses ORCA.
• Gaussian. A commercial quantum chemistry program, long the standard in many labs.
• Semiempirical methods (GFN2-xTB). Very fast approximate quantum methods, used for conformer searches and for large systems before a DFT calculation.
• Hartree-Fock (HF). The simplest wavefunction method. It lacks electron correlation and is rarely used alone today, although it is the standard for RESP charges.
• Kohn-Sham DFT. The practical form of DFT that uses a system of non interacting electrons with the same density as the real one.
• Born-Oppenheimer approximation. Treats the nuclei as fixed while the electrons are solved for, which is the basis of almost all quantum chemistry.
• QM/MM. Treats a small important region, such as a ligand and the residues around it, with quantum mechanics and the rest of the protein with a force field.

Cheminformatics and drug likeness

• SMILES. A line of text that describes a molecule, for example CCO for ethanol. It is the easiest way to give a ligand to a program. See the SMILES converter.
• InChI and InChIKey. A standard identifier of a molecule that is the same for everyone who draws it. The InChIKey is a short fixed length version for searching.
• SDF and MOL files. Standard formats for molecules with coordinates and bond orders. SDF can hold many molecules and their properties.
• MOL2. A format that holds coordinates, bond types, atom types and partial charges.
• PDB format. The standard text format for macromolecular structures from the Protein Data Bank. It has no bond orders for ligands unless CONECT records are given.
• Molecular descriptor. A number calculated from a structure, such as molecular weight, logP or the count of rotatable bonds. See RDKit descriptors.
• logP. The logarithm of the partition ratio between octanol and water. It measures lipophilicity, and drug like molecules usually have a logP below 5.
• Topological polar surface area (TPSA). The surface of polar atoms in a molecule. Values below about 140 Ų are linked to good absorption and below about 90 Ų to brain penetration.
• Hydrogen bond donors and acceptors. The counts of groups that give or accept hydrogen bonds. They affect solubility and membrane permeability.
• Lipinski rule of five. A guide for oral drugs. Molecular weight up to 500, logP up to 5, up to 5 hydrogen bond donors and up to 10 acceptors. A molecule with more than one violation is less likely to be orally available.
• Veber rules. A second guide for oral bioavailability. No more than 10 rotatable bonds and a TPSA of 140 Ų or less.
• PAINS. Pan assay interference compounds, substructures that often give false positives in screens and should be treated with care.
• Synthetic accessibility score. A number from 1 for easy to 10 for very hard that estimates how difficult a molecule is to make.
• QED. Quantitative estimate of drug likeness, a score from 0 to 1 that combines several descriptors.
• Molecular fingerprint. A bit string that encodes the structural features of a molecule, used to compare and search molecules.
• Tanimoto similarity. A number from 0 to 1 that compares two fingerprints. What a value means depends on the fingerprint, but with the common types a value above about 0.7 usually means closely related structures.
• Pharmacophore. The 3D arrangement of features, such as donors, acceptors, rings and charges, that a molecule needs to bind a target.
• QSAR. Quantitative structure activity relationship, a model that predicts activity from descriptors.
• ADMET. Absorption, distribution, metabolism, excretion and toxicity, the properties that decide whether an active molecule can become a drug.
• Scaffold. The core ring system of a series of compounds, kept fixed while the side groups change. ChemServer ligand design can lock a scaffold and mutate the rest.
• Bioisostere. A group that can replace another and keep similar biological activity, for example a tetrazole for a carboxylic acid.
• Conformer. One 3D shape of a flexible molecule obtained by rotating around single bonds. Docking and DFT both depend on starting from sensible conformers.
• RDKit. An open source cheminformatics toolkit used for descriptors, 3D generation and image drawing.
• Open Babel. An open source program that converts between more than 100 chemical file formats.
• Meeko. The program from the AutoDock team that prepares ligands and receptors as PDBQT files from RDKit molecules.