Nonlinear QSAR and 3D QSAR

Transcription

1 onlinear QSAR and 3D QSAR Hugo Kubinyi Germany HomePage onlinear Lipophilicity-Activity Relationships drug receptor Possible Reasons for onlinear Lipophilicity-Activity Relationships - kinetic control of drug transport in biological systems - equilibrium control of drug distribution - steric hindrance, allosteric effects - different pharmacokinetics and/or metabolism - solubility, micelle formation - end product inhibition of enzymes - drug receptor occupation

2 Parabolic Model (C. Hansch, 1964) log 1/C = a (log P) 2 + b log P + c log 1/C = a π 2 + b π + c log 1/C = a (log P) 2 + b log P + c σ + d log 1/C = a π 2 + b π + c σ + d E s + c, etc. Probability Model (J. McFarland, 1970) log 1/C = a log P - 2a log (P + 1) + c Equilibrium Model (R. Hyde, 1975) log 1/C = a log P - log (ap + 1) + c Bilinear Model (H. Kubinyi, 1976) log 1/C = a log P b log (ßP + 1) + c Hansch Multicompartment Model (1969) Definition P = k 1 /k 2 (correct) Hypothesis k 1.k 2 = 1 (wrong) 0 Log P

3 Hansch and Franke Models log 1/C = a log P + c (log P < log P x ) log 1/C = α (log P) 2 + ß log P + γ (log P > log P x ) McFarland Probability Model (1970) eurotoxic activity of alcohols, rat, i.p. application

4 Substance Distribution in a Three-Compartment System Rate constants of drug transport, k 1 and k 2 B. C. Lippold and G. F. Schneider, Arzneim.-Forsch. 25, 843 (1975) octanol phase B k 1 k 2 k 1 k 2 water phase A water phase C Homologous Quaternary -Alkylammoniumbromides, Water/n-ctanol/Water, + abr, experimental k 1 and k 2 values in h -1. umber of Carbon Atoms of Alkyl Group k 1 k

5 Bilinear Model log 1/C = a log P - b log (ßP + 1) + c Advantages of the bilinear model better fit of the linear left and right sides better description of the lipophilicity optimum Disadvantages of the bilinear model iterative estimation of the nonlinear parameter ß Loss of one degree of freedom (4 parameters, instead of 3)

6 eurotoxic Activity of Primary n-alcohols (rat, i.p. application) Log 1/C = (±0.05) log P (±0.10) log (ßP + 1) log ß = log P optimum = 1.94 (n = 10; r = 0.996; s = 0.041; F = 637.6) onlinear Lipophilicity-Activity Relationships

7 Dissociation und Ionization HA A - n-octanol P u P i HA + H 2 A - + H 3 + K a aqueous buffer Henderson-Hasselbalch Equation [H 3 + ] = K a.(f u /f i ) ph = pk a + log (f i /f u ) = pk a + log f i - log f u f u = fraction of neutral form (undissoziated acid or base) f i = α = fraction of charged form (= 1 f u ) Dissociation of Acids and Ionization of Bases P app [AH] org [AH] aq + [A - ] aq = (1 - α ) P u = P u ph - pk a Acids log P app = log P u - log ( ph - pk a ) P app [B] org [BH + ] aq + [B ] aq = (1 - α) P u = P u pk a - ph Bases log P app = log P u - log ( pk a - ph )

8 Distribution of Acids, Bases and eutral Compounds CH H 1, Salicylic Acid pk a = 3.0 H 3 C CH 3 CH 3 HCCH 3 Et 2, Phenacetin neutral 4, Caffeine pk a = 0.6 S CH 3 H 3 C CH 3 3, Promethazine pk a = 9.1 Approximation for bases for ph > pk a log P app = log P u for ph < pk a log P app log P u - pk a + ph for ph << pk a log P app = log P i

9 Phe Phe n-butyl 1, Phenylbutazone pk a = 4.5 Phe Et H CH 3 Phe Phe H H 2, Diphenylhydantoin pk a = 8.3 CH 3, -Methylphenobarbital CCH pk a = , Acetylsalicylic Acid pk a = 3.5 Approximation for acids for ph < pk a log P app = log P u for ph > pk a log P app log P u ph + pk a for ph >> pk a log P app = log P i

10 ph Dependence of the Absorption of Acids and Bases A strong base like strychnine, pk a = 8.3, is not toxic for a dog with a pylorus ligation. Gastrointestinal Absorption of eutral Compounds and Acids a) eutral drug b) Acid, pk a = 4 Stomach, ph = 1 Blood circulation, ph = 7.4 HA HA A - A - Stomach, ph = 1 Blood circulation, ph = 7.4 Intestine, ph = 6-8 HA A - HA A - Intestine, ph = 6-8

11 Gastrointestinal Absorption of Weak and Strong Bases c) Weak base, pk a = 5 d) Strong base, pk a = 9 B B BH + B Stomach, ph = 1 BH + BH + B BH + Stomach, ph = 1 Blood circulation, ph = 7.4 Blood circulation, ph = 7.4 B B B B BH + BH + Intestine, ph = 6-8 BH + BH + Intestine, ph = 6-8 Buccal Absorption of Acids and Bases Compound ph log P app log k abs Propranolol log P = pk a = p-hexylphenylacetic acid log P = 4.25 pk a = log k abs = 0.45 (±0.05) log P app (±0.05) log ( P app + 1) (n = 12; r = 0.988; s = 0.102)

12 The ph Shift in the Absorption of Lipophilic Acids and Bases Distributed and absorbed amount of acid HA ph = ph absorption profile (kinetic control) ph partitioning profile (equilibrium) ph shift ph value 3D-QSAR Approaches CoMFA, Comparative Molecular Field Analysis (Richard Cramer et al., 1988) Select training and test sets of comparable diversity. Generate 3D structures of all molecules of the data set. Establish orientation rules for superposition of the molecules, using e.g. the "active analog approach". Align the molecules according to their pharmacophore and surface properties, using e.g. the program SEAL. Insert the molecules in a box and generate a grid that covers also a sufficiently large volume around the molecules.

13 Box and Grid CH CH 3 3 R Calculate property fields for every molecule, in each grid point, by using probe atoms or groups (steric and electrostatic effects are most often separately treated; in addition, hydrophobic effects as well as hydrogen bond donor and acceptor properties may be considered). Grid points with low variance may be neglected or variable / region selection can be performed. Perform PLS analysis, using an optimum number of latent variables (vectors) to correlate biological activities. Check the internal predictivity of the PLS model by a stepwise elimination of objects (crossvalidation). If necessary, repeat these steps. Predict the biological activities of the test set molecules.

14 Electrostatic, Steric and Similarity Fields Coulomb potential E C = n q q i i= 1 Dr j ij Lennard-Jones potential n vdw j ij 12 E = A r C j rij 6 i= 1 ( ) SEAL similarity coefficients m n AF wije r 2 = α ij ; w i = 1j = 1 ij = w E q i q j + w S v i v j +... E(r) Van-der-Waals potential, CoMFA (Lennard-Jones function) cut-off value CoMFA and CoMSIA Potentials Gaussian approximation, CoMSIA Coulomb potential, CoMFA (identical charges) 0 r

15 PLS (Partial Least Squares) Analysis X 1 From u = kt (k = constants k 1, k 2... k j ; j = number of PLS vektors) follows X 2 X 3 BA i = a 1 S i1 + a 2 S i2 + a 3 S i a m S im + b 1 E i1 + b 2 E i2 + b 3 E i b m E im Y vector BA 1 BA 2 BA 3 BA n S 11 S 12 S 13 S 21 S 22 S 23 S 31 S n1 S 32 S n2 X matrix S 33 S n3 S S S S n S 1m E 11 S 2m E 21 S 3 m S nm E 31 E n1 E E E E n E 1m E 2m E 3m E nm PLS analysis SAMPLS analysis u 11 u 12 u 13 t 11 t 12 t 13 u 1n,,... t 1n t 21 t 22 t 23 t 2n PLS vectors (latent variables) c 11 c 12 c c 1n c 21 c 22 c c 2n c 31 c 32 c c 3n c n1 c n2 c n c nn covariance matrix

16 Steric CoMFA Contour Map Electrostatic CoMFA Contour Map

17 Recommendations for CoMFA Analyses ( good practice ) Type of biological data (affinities, inhibition constants) Variance and error range of biological data Selection of start geometries (flexible molecules) Method for calculation of charges should be cited Pharmacophore for superposition of the molecules Description of the alignment (atom-by-atom, field-based) Scaling and weighting of fields umber of PLS vectors ( ccam s razor ) Variable or region selection Crossvalidation - L or groups Crossvalidation only for internal prediction - Prediction of a test set! QSAR and 3D-QSAR, Scope and Limitations a) Free Wilson Analysis + easy to perform, most often unique solutions + clear separation of substituent effects + helps in the derivation of Hansch models + combination with Hansch analysis to a mixed approach - needs at least two sites of chemical variation - many parameters, only few degrees of freedom - very narrow QSAR models, no "outside" predictions

18 b) Hansch Analysis + correlates activities with physicochemical properties + "outside" predictions are possible - only applicable to congeneric series - works best with simple variation of aromatic substituents - considers only 2D structures - no unique solutions - risk of chance correlations (number of variables!) - risk of failure in "too far outside" predictions c) 3D QSAR (CoMFA and CoMSIA) + considers 3D structures of the ligands + applicable to more heterogeneous data sets + elektrostatic, steric, hydrophobic and hydrogen bond fields + 3D maps of favorable and unfavorable interactions - uncertainties about the bioactive conformation - uncertainties about different binding modes of ligands - cut-off levels (can be avoided in CoMSIA) - variable selection yields fragmented contour maps - high risk (if not guarantee) of chance correlations - only applicable to in vitro data