GLOSSARY

application data folders: Effective with this release, personalizable model definition and support files can be found in:

C:\Users\<user_name>\AppData\Local\Simulations Plus, Inc\<app>

where "<user_name>" is your user name and "<app>" is the version number of the program of interest, i.e., ADMET_Predictor9.5 or ADMET_Modeler9.5. The latter includes a \Temp directory where work in progress is stored. Note that the AppData folder is hidden by default, so navigating to it may require you to copy and paste the path into the Windows Open dialog and provide the user and application names. See also: installation directories.

aqueous solubility: the amount of a solute that can dissolve per volume of water — depends on pH for compounds with acidic or basic substructures. It also depends on ionic strength and temperature.

attribute groups are indicated on the display tabs positioned along the bottom the ADMET Predictor spreadsheet to allow you to control which attributes are displayed at any time. The base groups for ADMET properties are named in parallel with the licensed modules - PhysChem, Transporters, Metabolism and Toxicity for the Physicochemical and Biopharmaceutical, Transporters, Metabolism and Toxicity Modules, respectively. The User Inputs, Descriptors and User Models attribute groups are accessible so long as a valid license exists for any module. The Risks group includes ADMET Risk and the more specific Risk models that contribute to it. Users can define their own attribute sets using the Customize button at the end of the row. a model group is a special case of an attribute group.

CYP: Cytochrome P450s are oxidative enzymes that participate mostly in Phase I metabolism.

concordance: fraction of predictions that are correct.

diffusion coefficient, diffusivity: the rate of diffusion of a substance along a concentration gradient.

effective permeability (Peff): quantifies permeability across the intestinal membrane. Peff depends on properties of both the drug and the membrane. It is referred to as “effective” because it characterizes absorption whether by passive diffusion, active transport, facilitated diffusion, paracellular diffusion, or any other mechanism. Intestinal Peff is measured in man by regional jejunal perfusion. Inlet concentrations (C_in) and outlet concentrations (C_out) of solute are measured from solution perfused at a fixed flow rate (Q). Area of the mass-transfer surface is 2RL, where R and L are the radius and length of the jejunal segment. At steady state:

P_eff = (C_in - C_out)×Q/(2×R ×L×C_in)

with units of L per time. (Lennernas 1992).

generic name: the common name for the active ingredient in a drug.

hERG: human Ether-à-go-go Related Gene product, a potassium channel responsible for the repolarization of the cardiac action potential.

intestinal absorption: intestinal absorption refers to the process of a drug being absorbed across the intestinal membrane. It does not refer to the complete process of a drug being absorbed, passing through the liver, and arriving in the systemic circulation. Intestinal absorption can occur through several different processes: 1) active transport, 2) facilitated diffusion, 3) passive diffusion, and 4) paracellular diffusion.

intrinsic solubility: equilibrium concentration of the neutral (un-ionized) form of a compound in solution at saturation with respect to a neutral solid in pure water.

installation directories and installation folder: The Windows folder where ADMET Predictor is installed. For 32 bit executables, this is in:

C:\Program Files (x86)\Simulations Plus, Inc\ADMET_Predictor<v>

where "<v>" indicates the version number (e.g., "9.5"); 64 bit executables go into the Program Files directory rather than into Program Files (x86). Model definition and support files are stored in a separate [application folder ]specific to each user.

ionizable group: a molecular substructure that can gain a hydrogen ion from or lose a hydrogen ion to another molecule with a concomitant increase in formal charge. Interactions between ionizable groups determine the pK_a values for a molecule.

J: see Youden's index.

Log P: the common logarithm of the partition coefficient P between 1-octanol and water. P is the ratio of activities (often approximated by concentrations) of a solute distributed between octanol and water. The more positive a compound's log P is, the more hydrophobic (and lipophilic) it is. Hydrophilic compounds, in contrast, have more negative log P values.

logit transform: f(y) = ln(y/(1-y)) applied on the interval (0,1). The inverse function ("antilogit") is g(y) = 1/(1+exp(-x)) - a special cased of the logistic function.

macrostate, macrospecies: a collection of microstates characterized by the same net charge

macroscopic pKa: the negative log of the equilibrium dissociation constant between two consecutive macrostates of a molecule.

MAE: mean absolute error.

MCC: generalization of the binary Matthews correlation coefficient, as described by Gorodkin (2004).

MDCK: Madin-Darby Canine Kidney. The cell line bearing this name is widely used to assess permeability of drug candidates in vitro (Irvine 1999)

microstate, microspecies: state of an ionizable molecule with protonation states specified for every ionizable center.

MLR: multiple linear regression.

model group: An attribute group comprised of models. Most models are only enabled by one Module license, but some are enabled by more than one license. As a result, the mapping from Model Groups to Modules is not one-to-one.

MRTD: maximum recommended therapeutic dose.

native pH: The equilibrium pH at saturation when a compound in its neutral solid state is dissolved in pure water.

native solubility: equilibrium concentration at saturation for a neutral form of a pure chemical compound dissolved in pure water (pure water solubility at the native pH). Note: unlike intrinsic solubility which is defined by chemical equilibrium between neutral species in solid state and in solution, the native solubility concerns all forms (neutral and ionized) in solution.

Peff: see effective permeability.

Q-squared: coefficient of determination.

R-squared: the square of Pearson's correlation coefficient; equals the fraction of the variance accounted for by the model.

RMSE: root mean square error of regression.

RMSU: root mean squared uncertainty.

sensitivity: fraction of positive observations that are classified correctly by a model.

solubility factor: ratio of the equilibrium solubility of an ionizable chemical compound’s salt to its intrinsic solubility (parameter specific for each salt).

solubility profile: a set of points on the pH-concentration graph; each point represents thermodynamic equilibrium solubility at given pH. Note: the exact solid form (salt, neutral) of a compound under equilibrium solubility conditions may be different in different pH regions.

SMILES structure: SMILES is an acronym for Simplified Molecular Input Line Entry Specification and is used to encode a particular valence model of a molecular structure as a linear string (http://www.daylight.com/smiles/index.html). The SMILES code for nitroethane, for example, is CC[N+](=O)[O-].

specificity: fraction of negative observations that are classified correctly by a model.

SRCC: Spearman's rank correlation coefficient.

systematic name: the IUPAC name of the drug.

UGT: Uridine diphosphate glucuronosyltransferases are conjugating enzymes that primarily participate in Phase II metabolism..

user inputs: attributes read directly from input file(s).

Vd: Volume of distribution.

Youden's index (J): For binary classifications, J = Sensitivity+Specificity-1 (Youden 1950). For non-binary classification models, we have generalized this index as follows:

where the sums are over all k classes, N is the total number of observations, and N_correct is the total number of observations correctly classified. o_i and p_i represent the number of observations in class i and the number of observations predicted to be in class i, respectively. It can be shown that this reduces to the Youden index in the case of 2 classes. For multiple classes, this generalization has the property (as does the binary form) that it ranges between -1 and 1, with a perfect result yielding 1, a random result yielding 0, and a worse than random result yielding a negative number.