QSAR Methodology

Traditional Hansch analysis using multiple linear regression (MLR) suffers from several shortcomings. One of the problems is that one often has more variables than compounds and Hansch analysis ideally requires at least five times as many compounds as descriptors. Hansch analysis becomes unstable when correlated descriptors are used. Furthermore, there is often a need to consider correlations between chemical descriptors and several biological tests simultaneously. 

A good example of a comparative study using Hansch and FW, both using MLR, with a PLS analysis on the same data, is given in.  

Progress in chemometrics has made a number of new statistical techniques available, which are inaeasingly being used. This concerns both new supervised and unsupervised (or pattern recognition ) techniques. Chemometrics was defined about 25 years ago as the chemical discipline which uses mathematical, statistical and related techniques to design optimal measurement procedures and experiments, and to extract maximum relevant information from chemical data. The science of chemometrics has been developed to promote applications of statistics in analytical, organic and medicinal chemistry. 

The current focus on early prediction of ADMET properties, as well as the analysis of HTS data, has led to a revival and extension of the use of QSAR technology in the pharmaceutical industry. Other applications can be found in metabonomics, the application of chemometrics to analytical spectral data to predict disease or the effect of compounds on metabolism. Many methods originally known in economics and artificial intelligence research are now also being used in QSAR/QSPR. 

We see currently a change from academic QSAR, with models using 10-100 s of compounds, to industrial QSAR, 

---

The QSPR/QSAR methodology can also be applied to materials and mixtures where no structural information is available. Instead of descriptors derived from the compound s structure, various physicochemical properties, including spectra, can be used. In particular, spectra are valuable in this context as they reflect the structure in a sensitive way. 

The US EPA T.E.S.T. is a downloadable program to estimate different toxicological endpoints and physicochemical properties from molecular structure using a variety of QSAR methodologies [58],... 

Dick Cramer provided insight and inspirahon that led to my interest in 3D QSAR methodology ]40] and was the impetus (the precursor of CoMFA was a lattice model [41] developed by Cramer and Milne at SKF) behind the development of CoMFA (Comparative Molecular Field Analysis) by Tripos [42], The success of CoMFA in... 

Gussio, R., Pattabieaman, N., Kellogg, G.E., Zahaeevitz, D.W. Use of 3D QSAR methodology for data mining the National Cancer Institute Repository of Small Molecules application to HlV-1 reverse transcriptase inhibition. Methods 1998, 14, 255-263. 

The popularity of commercial programs such as Comparative Molecular Field Analysis (4,12) (CoMFA) and Catalyst (13) has limited both the evaluation and use of other QSAR methodologies. Often well-known issues associated with CoMFA and Catalyst have come to be viewed as shortcomings that simply are accepted as working limitations in a 3D-QSAR analysis. In this section we challenge this position and present 3D- and nD-QSAR methods that are able to overcome some of the issues associated with current mainstream 3D-QSAR application products. 

The methodology of nD-QSAR adds to the 3D-QSAR methodology by incorporating unique physical characteristics, or a set of characteristics, to the descriptor pool available for the creation of the models. The methods of Eigenvalue Analysis (40) (EVA) and 4D-QSAR (5) are examples of using unique physical characteristics in the creation of a QSAR model. 4D-QSAR uses an ensemble of molecular conformations to aid in the creation of a QSAR. The EVA-QSAR method uses infrared spectra to extract descriptors for the creation of the QSAR model. 

Although ADMET-QSAR methodologies are traditionally not considered RD methods, a pseudo-receptor may be needed to extract relevant ADMET descrip-... 

Traditional and 3D-QSAR methodologies have three inherent limitations ... 

Constructing the model using all the calculated descriptors is feasible, but this leads to overgeneralized models. The problem of many descriptors is further compounded when using 3D and /zD-QSAR methodology where thousands of descriptors are calculated. The end results of calculating all possible descriptors for traditional QSAR and the data returned from 3D-QSAR methods are tables containing many columns (descriptors) and few rows (molecules, bioactivities). 

Is the QSAR methodology appropriate for the desired use of the QSAR model ... 

Ekins et al. (201) used the MS-WHIM descriptors to construct 3D and 4D QSAR models for the log(l/Aj) of 14 competitive inhibitors of CYP3A. The 3D QSAR of the CYP3A4-mediated midazolam l -hydroxylation was shown to be predictive yielding a leave-one-out (LOO) q2 value of 0.32. Although the 4D QSAR methodology includes conformational changes, it did not provide for a significant improvement over the 3D QSAR (LOO q2 0.44). Two other datasets (242,243) were used to create 3D and 4D QSAR models. In both datasets, it was not possible to build predictive 3D QSAR models however, 4D QSAR models were constructed (LOO q2 = 0.41-0.56). 

QSAR modeling has been traditionally viewed as an evaluative approach, i.e., with the focus on developing retrospective and explanatory models of existing data. Model extrapolation has been considered only in hypothetical sense in terms of potential modifications of known biologically active chemicals that could improve compounds activity. Nevertheless recent studies suggest that current QSAR methodologies may afford robust and validated models capable of accurate prediction of compound properties for molecules not included in the training sets. 

Varnek of Universite Louis Pasteur gives a comprehensive explanation of QSAR methodology (see Chapter 5 of this volume). 

There is the need of comparison and integration of different QSAR methodologies. The regulatory needs forced the debate on the comparison of the models. New efforts have to be put in this direction because there are many open problems. The components of the QSAR models are still in many cases not fully clarified. 

---