Ghose and Crippen

Method of Ghose and Crippen The method of Ghose and Crippen [19] uses 120 different atom types. They are described for the corresponding Kow model in Chapter 12. A training set of 538 compounds was employed. Observed versus calculated R d showed a correlation coefficient of 0.998 and a standard deviation of 

---

Log P calculation for quinidine with the atom contribution method according to Ghose and Crippen. R group connected to C X heteroatom = double bond aromatic bond aromatic single bond (e.g. C=N in pyrrole) subscripts give the hybridization state and superscripts the formal oxidization number. For the quinidine structure see Fig. 14.1. 

The function fct d) has no physical basis. According to Heiden (6), it should fulfill only two conditions it should be smooth and continuous and have finite values for d < dcut oS. The value of dclH 0[f should be larger than the van der Waals radius of any atom in the molecule under consideration. In the program LipoDyn, the Ghose and Crippen parameters (18) as well as the Broto and Moreau parameter set (16) are implemented, with the the exponential function used by Fauchere (cxp(-d)) (3), the hyperbolic function defined by Audry 1/(1 + d) (2) and the parameterized Fermi distance function used by Testa and coworkers (1). 

One of the limitations of current log Poct prediction techniques is in the accuracy of the atomic fragmental system used. The weaknesses of the atomic fragmentation systems of Ghose and Crippen or of Broto and Moreau have been... 

Fig. 5. Estimation of log Poct using MLP calculation with the atomic fragmental system of Ghose and Crippen and the Fauchere distance function on the 91 selected WDI rigid molecules.

Figure 4.4.1 Estimation of R and n(,° for 6-methyl-5-heptene-2-one using the method of Ghose and Crippen [21].

Figure 13.4.5 Estimation of Kow for 2-bromoethyl ethanoate using the method of Ghose and Crippen [48] (constants from [50]).

Based on this in-house dataset, an in-silico prediction model [27] (three-layered neural network, Ghose and Crippen [28,29] descriptors) was constructed to evaluate the frequent hitter potential before compound libraries are purchased or synthesized. This model was validated with a dataset of the above-mentioned promiscuous ligands published by McGovern et al. [26], in which 25 out of 31 compounds were correctly recognized. 

Figure 12.2 The compound sets from which the hERG (filled triangles) and the BBB (open circles) in-silico filters were derived are compared by principal components analysis, and one structure of each set is depicted. Ghose and Crippen descriptors were calculated for all the molecules, and after autoscaling, the compounds were projected onto the scores plot of the first two components [74],...

Lipophilicity descriptors, in particular, the atomic lipophilicity contribution La in Ghose and Crippen s system, taking into account the environment of an atom, and group lipophilicity Lg defined as a sum of contributions for a non-hydrogen atom and attached hydrogens. 

Artificial Neural Networks and Decision Trees. Figure 6.3 shows an example of a simple neural network that uses Ghose and Crippen atom types (43)to code the molecular... 

The Ghose-Crippen contribution method is based on hydrophobic atomic constants a measuring the lipophilic contributions of atoms in the molecule, each described by its neighbouring atoms [Ghose and Crippen, 1986 Ghose et al, 1988 Viswanadhan et al, 1989]. 

Values for the atomic molar refractivity were also estimated by - group contribution methods [Ghose and Crippen, 1987],... 

Other atom-centered fragments were proposed on the basis of different sets of rules and mainly used in —> group contrifowtion methods for estimation of molecular properties such as lipophilicity and —> molar refractivity [Ghose and Crippen, 1986 Viswanadhan, Ghose et al, 1989 Mekenyan, Bonchev et al, 1987 Meylan, Howard et al, 1992 Lohninger, 1994 Baumann and Glerc, 1997 Wildman and Crippen, 1999]. 

Ghose and Crippen [89] first worked on this approach on a more theoretical basis. Later, Smellie et al. [90] applied this methodology to real test cases. Billeter et al. [91] combined the description of molecules by distance constraints with an efficient algorithm for constrained optimization. 

---